DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
This action is in response to amendments and remarks filed on 07/16/2026. In the current amendments, claims 1, 6-7, and 15-16 are amended. Claims 1-20 are pending and have been examined.
Claim Objections
Claims 1-20 are objected to because of the following informalities:
In claim 1, line 5, “categories according to feature” should read “categories according to a feature”
In claim 6, line 7, “a computing circuit and a on-chip static random access memory (SRAM)” should read “a computing circuit and an on-chip static random access memory (SRAM)”
In claim 6, line 9, “making the computing circuit to execute” should read “making the computing circuit execute”
In claim 6, line 16, “categories according to feature” should read “categories according to a feature”
In claim 15, lines 6-7, “a computing circuit and a on-chip static random access memory (SRAM)” should read “a computing circuit and an on-chip static random access memory (SRAM)”
In claim 15, line 8, “making the computing circuit to execute” should read “making the computing circuit execute”
In claim 15, line 11, “according to feature” should read “according to a feature”
Dependent claims 2-5 are objected based on being directly or indirectly dependent on objected claim 1.
Dependent claims 7-15 are objected based on being directly or indirectly dependent on objected claim 6.
Dependent claims 16-20 are objected based on being directly or indirectly dependent on objected claim 15.
Appropriate correction is required.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 1-5 and 15-20 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claim 1 recites the limitation “the corresponding model” in lines 17-18. There is insufficient antecedent basis for this limitation in the claim. For examination purposes, “the corresponding model” has been interpreted as “a corresponding model”.
Claim 1 recites the limitation “the classification result” in line 18. There is insufficient antecedent basis for this limitation in the claim. For examination purposes, “the classification result” has been interpreted as “a classification result”.
Claim 1 recites the limitation “the data” in line 19. There is insufficient antecedent basis for this limitation in the claim. For examination purposes, “the data” has been interpreted as “data”.
Claim 15 recites the limitation “the processor” in line 11. There is insufficient antecedent basis for this limitation in the claim. For examination purposes, “the processor” has been interpreted as “a processor”.
Dependent claims 2-5 are rejected based on being directly or indirectly dependent on rejected claim 1.
Dependent claims 16-20 are rejected based on being directly or indirectly dependent on rejected claim 15.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding Claim 1,
Claim 1 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 Analysis: Claim 1 is directed to a method, which is directed to a process, one of the statutory categories.
Step 2A Prong One Analysis: The limitations:
“establishing a plurality of categories according to feature of each of a plurality of classes by the processor, wherein each of the plurality of categories includes at least one of the plurality of classes with same features”
“using at least one first class within one of the plurality of categories for verification”
As drafted, under their broadest reasonable interpretations, cover mental processes (concepts performed in the human mind (including an observation, evaluation, judgement, opinion)) but for the recitation of mere instructions to apply language (See MPEP 2106.05(f)) and insignificant extra-solution activity (See MPEP 2106.05(g)). The above limitations in the context of this claim encompass establishing categories according to a feature of each of a plurality of classes, the categories including classes with same features (corresponds to evaluation and judgement; in particular, a human, with the assistance of pen and paper, can establish categories of classes with same features); and using the first class within the first category for verification (corresponds to evaluation and judgement; in particular, a human, with the assistance of pen and paper, can perform verification based on the first class within the first category).
Step 2A Prong Two Analysis: The judicial exceptions are not integrated into a practical application. In particular, the claim recites additional elements that are mere instructions to apply (See MPEP 2106.05(f)) or insignificant extra-solution activity (See MPEP 2106.05(g)). The limitations:
“an accelerator”
“a processor”
“training a first model with the training dataset labeled with the plurality of classes”
“by the processor”
“disposing a computing circuit and a memory on a same chip to form the accelerator”
“implementing the first model on the accelerator by the computing circuit and the memory”
“accessing the memory internally in the chip in order to execute the corresponding model by the computing circuit for generating the classification result upon receiving of the data”
As drafted, are additional elements that amount to no more than mere instructions to apply the exception for the abstract ideas. See MPEP 2106.05(f). The limitations:
“receiving a training dataset having a plurality of training data by a processor, wherein each training data is labeled to one of a plurality of classes”
As drafted, are additional elements that correspond to insignificant extra-solution activity. In particular, the additional elements are merely directed towards mere data gathering. See MPEP 2106.05(g). Therefore, the additional elements do not integrate the abstract ideas into a practical application.
Step 2B Analysis: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, all of the additional elements are “mere instructions to apply an exception” (I.e. the additional elements describe a generic accelerator, processor, memory, computing circuit, model, and generic training of the model for applying the abstract ideas) or insignificant extra-solution activity (i.e. receiving data). Furthermore, the “receiving …” limitation is insignificant extra-solution activity that is well-understood, routine, and conventional according to MPEP 2106.05(d) (“The courts have recognized the following computer functions as well‐understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity… i. Receiving or transmitting data over a network). Mere instructions to apply an exception cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 2,
Claim 2 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 Analysis: Claim 2 is directed to a method, which is directed to a process, one of the statutory categories.
Step 2A Prong One Analysis: The limitations:
“generate at least one first probability value, respectively corresponding to the at least one first class, for inferring percentages an object of the at least one class is shown in each training data”
As drafted, under their broadest reasonable interpretations, cover mental processes (concepts performed in the human mind (including an observation, evaluation, judgement, opinion)) but for the recitation of mere instructions to apply language (See MPEP 2106.05(f)) and insignificant extra-solution activity (See MPEP 2106.05(g)). The above limitations in the context of this claim encompass generating a first probability value corresponding to the first class to infer the percentage of an object of the first class being in each training data (corresponds to evaluation and judgement; in particular, a human, with the assistance of pen and paper, can generate a probability for inferring percentages of an object of the first class being shown in each training data).
Step 2A Prong Two Analysis: The judicial exceptions are not integrated into a practical application. In particular, the claim recites additional elements that are mere instructions to apply (See MPEP 2106.05(f)) or insignificant extra-solution activity (See MPEP 2106.05(g)). The limitations:
“the trained first model”
As drafted, are additional elements that amount to no more than mere instructions to apply the exception for the abstract ideas. See MPEP 2106.05(f). In addition, the recitation of additional elements in claim 1 of a generic accelerator, processor, memory, computing circuit, model, and generic training of the model, as drafted, are reciting mere instructions to apply language such that it amounts to no more than mere instructions to apply the exceptions. In addition, the additional element of “receiving …” amounts to no more than insignificant extra-solution activity for receiving data. Therefore, the additional elements do not integrate the abstract ideas into a practical application.
Step 2B Analysis: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, all of the additional elements are “mere instructions to apply an exception” (I.e. the additional elements describe a generic accelerator, processor, memory, computing circuit, model, and generic training of the model for applying the abstract ideas) or insignificant extra-solution activity (i.e. receiving data). Furthermore, the “receiving …” limitation is insignificant extra-solution activity that is well-understood, routine, and conventional according to MPEP 2106.05(d) (“The courts have recognized the following computer functions as well‐understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity… i. Receiving or transmitting data over a network). Mere instructions to apply an exception cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 3,
Claim 3 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 Analysis: Claim 3 is directed to a method, which is directed to a process, one of the statutory categories.
Step 2A Prong One Analysis: The limitations:
“generate a first category probability value, for inferring a percentage whether objects of the first category is shown in each training data”
As drafted, under their broadest reasonable interpretations, cover mental processes (concepts performed in the human mind (including an observation, evaluation, judgement, opinion)) but for the recitation of mere instructions to apply language (See MPEP 2106.05(f)) and insignificant extra-solution activity (See MPEP 2106.05(g)). The above limitations in the context of this claim encompass generating a first category probability value to infer the percentage of an object of the first category being in each training data (corresponds to evaluation and judgement; in particular, a human, with the assistance of pen and paper, can generate a category probability for inferring percentages of an object of the first category being shown in each training data).
Step 2A Prong Two Analysis: The judicial exceptions are not integrated into a practical application. In particular, the claim recites additional elements that are mere instructions to apply (See MPEP 2106.05(f)) or insignificant extra-solution activity (See MPEP 2106.05(g)). The limitations:
“the trained first model”
As drafted, are additional elements that amount to no more than mere instructions to apply the exception for the abstract ideas. See MPEP 2106.05(f). In addition, the recitation of additional elements in claim 2 of a generic accelerator, processor, memory, computing circuit, model, and generic training of the model, as drafted, are reciting mere instructions to apply language such that it amounts to no more than mere instructions to apply the exceptions. In addition, the additional element of “receiving …” amounts to no more than insignificant extra-solution activity for receiving data. Therefore, the additional elements do not integrate the abstract ideas into a practical application.
Step 2B Analysis: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, all of the additional elements are “mere instructions to apply an exception” (I.e. the additional elements describe a generic accelerator, processor, memory, computing circuit, model, and generic training of the model for applying the abstract ideas) or insignificant extra-solution activity (i.e. receiving data). Furthermore, the “receiving …” limitation is insignificant extra-solution activity that is well-understood, routine, and conventional according to MPEP 2106.05(d) (“The courts have recognized the following computer functions as well‐understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity… i. Receiving or transmitting data over a network). Mere instructions to apply an exception cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 4,
Claim 4 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 Analysis: Claim 4 is directed to a method, which is directed to a process, one of the statutory categories.
Step 2A Prong One Analysis: The limitations:
“wherein a summation of the first category probability value and the at least one first probability value equals to 1”
As drafted, is part of the abstract ideas of claim 3 of generating a first probability value and generating a first category probability value. The limitation of claim 4 further limits the limitations of claim 3 by further defining the probability values. The above limitation in the context of this claim encompasses generating a first probability value and generating a first category probability value, wherein the summation of the probability values is equal to 1 (corresponds to evaluation and judgement; in particular, a human, with the assistance of pen and paper, can generate a first category probability value and generate at least one first probability value, with the summation of the probability values being 1).
Step 2A Prong Two Analysis: The judicial exceptions are not integrated into a practical application. In particular, the claim recites additional elements that are mere instructions to apply (See MPEP 2106.05(f)) or insignificant extra-solution activity (See MPEP 2106.05(g)). The recitation of additional elements in claim 3 of a generic accelerator, processor, memory, computing circuit, model, and generic training of the model, as drafted, are reciting mere instructions to apply language such that it amounts to no more than mere instructions to apply the exceptions. In addition, the additional element of “receiving …” amounts to no more than insignificant extra-solution activity for receiving data. Therefore, the additional elements do not integrate the abstract ideas into a practical application.
Step 2B Analysis: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, all of the additional elements are “mere instructions to apply an exception” (I.e. the additional elements describe a generic accelerator, processor, memory, computing circuit, model, and generic training of the model for applying the abstract ideas) or insignificant extra-solution activity (i.e. receiving data). Furthermore, the “receiving …” limitation is insignificant extra-solution activity that is well-understood, routine, and conventional according to MPEP 2106.05(d) (“The courts have recognized the following computer functions as well‐understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity… i. Receiving or transmitting data over a network). Mere instructions to apply an exception cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 5,
Claim 5 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 Analysis: Claim 5 is directed to a method, which is directed to a process, one of the statutory categories.
Step 2A Prong One Analysis: The limitations:
“establishing a first category class by merging all classes that fall outside of the first category”
“verifying the first model by using the first category class and the at least one first class”
As drafted, under their broadest reasonable interpretations, cover mental processes (concepts performed in the human mind (including an observation, evaluation, judgement, opinion)) but for the recitation of mere instructions to apply language (See MPEP 2106.05(f)) and insignificant extra-solution activity (See MPEP 2106.05(g)). The above limitations in the context of this claim encompass establishing a first category class by merging all classes that fall outside of the first category (corresponds to evaluation and judgement; in particular, a human, with the assistance of pen and paper, can merge all classes outside of the first category to merge a first category class); and verifying the first model by using the first category class and the first class (corresponds to evaluation and judgement; in particular, a human, with the assistance of pen and paper, can use the first category class and the first class to verify the first model).
Step 2A Prong Two Analysis: The judicial exceptions are not integrated into a practical application. In particular, the claim recites additional elements that are mere instructions to apply (See MPEP 2106.05(f)) or insignificant extra-solution activity (See MPEP 2106.05(g)). The limitations:
“training the first model with the training dataset”
As drafted, are additional elements that amount to no more than mere instructions to apply the exception for the abstract ideas. See MPEP 2106.05(f). In addition, the recitation of additional elements in claim 1 of a generic accelerator, processor, memory, computing circuit, model, and generic training of the model, as drafted, are reciting mere instructions to apply language such that it amounts to no more than mere instructions to apply the exceptions. In addition, the additional element of “receiving …” amounts to no more than insignificant extra-solution activity for receiving data. Therefore, the additional elements do not integrate the abstract ideas into a practical application.
Step 2B Analysis: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, all of the additional elements are “mere instructions to apply an exception” (I.e. the additional elements describe a generic accelerator, processor, memory, computing circuit, model, and generic training of the model for applying the abstract ideas) or insignificant extra-solution activity (i.e. receiving data). Furthermore, the “receiving …” limitation is insignificant extra-solution activity that is well-understood, routine, and conventional according to MPEP 2106.05(d) (“The courts have recognized the following computer functions as well‐understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity… i. Receiving or transmitting data over a network). Mere instructions to apply an exception cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 6,
Claim 6 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 Analysis: Claim 6 is directed to an electronic system, which is directed to a machine, one of the statutory categories.
Step 2A Prong One Analysis: The limitations:
“upon receiving of a data, … generating a classification result comprising at least one probability value to infer whether the data falls within the corresponding category”
“establish the plurality of categories according to feature of each of a plurality of classes, wherein each of the plurality of categories includes at least one of the plurality of classes with same features”
As drafted, under their broadest reasonable interpretations, cover mental processes (concepts performed in the human mind (including an observation, evaluation, judgement, opinion)) but for the recitation of mere instructions to apply language (See MPEP 2106.05(f)) and insignificant extra-solution activity (See MPEP 2106.05(g)). The above limitations in the context of this claim encompass generating a classification result comprising a probability value to infer if data is part of the corresponding category (corresponds to evaluation and judgement; in particular, a human, with the assistance of pen and paper, can generate a classification comprising a probability value to determine if data falls within the corresponding category); and establishing categories according to a feature of each of a plurality of classes, the categories including classes with same features (corresponds to evaluation and judgement; in particular, a human, with the assistance of pen and paper, can establish categories of classes with same features).
Step 2A Prong Two Analysis: The judicial exceptions are not integrated into a practical application. In particular, the claim recites additional elements that are mere instructions to apply (See MPEP 2106.05(f)) or insignificant extra-solution activity (See MPEP 2106.05(g)). The limitations:
“a processor”
“a plurality of accelerators, coupled to the processor”
“each accelerator”
“wherein the model is trained using a training dataset labeled with the at least one class”
“wherein each accelerator comprises a computing circuit and a on-chip static random access memory (SRAM)”
“making the computing circuit to execute the model”
“wherein the plurality of accelerators and corresponding SRAMs are respectively disposed on a plurality of chips, and the plurality of accelerators are respectively corresponding to the plurality of categories”
“the processor”
As drafted, are additional elements that amount to no more than mere instructions to apply the exception for the abstract ideas. See MPEP 2106.05(f). The limitations:
“store a model corresponding to one of a plurality of categories with at least one class having a same feature and being categorized within the category”
As drafted, are additional elements that correspond to insignificant extra-solution activity. In particular, the additional elements are merely directed towards mere data gathering. See MPEP 2106.05(g). Therefore, the additional elements do not integrate the abstract ideas into a practical application.
Step 2B Analysis: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, all of the additional elements are “mere instructions to apply an exception” (I.e. the additional elements describe a generic processor, accelerators, computing circuit, SRAM, model, and generic training of the model for applying the abstract ideas) or insignificant extra-solution activity (i.e. storing data). Furthermore, the “store …” limitation is insignificant extra-solution activity that is well-understood, routine, and conventional according to MPEP 2106.05(d) (“The courts have recognized the following computer functions as well‐understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity… iv. Storing and retrieving information in memory). Mere instructions to apply an exception cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 7,
Claim 7 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 Analysis: Claim 7 is directed to an electronic system, which is directed to a machine, one of the statutory categories.
Step 2A Prong One Analysis: Please see the analysis of claim 6. The limitation of claim 7 is only an additional element to the abstract ideas of claim 6.
Step 2A Prong Two Analysis: The judicial exceptions are not integrated into a practical application. In particular, the claim recites additional elements that are mere instructions to apply (See MPEP 2106.05(f)) or insignificant extra-solution activity (See MPEP 2106.05(g)). The limitations:
“each of the plurality of accelerators”
“the SRAM”
“the computing circuit, coupled to the SRAM”
As drafted, are additional elements that amount to no more than mere instructions to apply the exception for the abstract ideas. See MPEP 2106.05(f). The limitations:
“store the corresponding model”
“access the SRAM on the same chip internally in order to execute the corresponding model for generating the classification result upon receiving of the data”
As drafted, are additional elements that correspond to insignificant extra-solution activity. In particular, the additional elements are merely directed towards mere data gathering. See MPEP 2106.05(g). In addition, the recitation of additional elements in claim 6 of a generic processor, accelerators, computing circuit, SRAM, model, and generic training of the model, as drafted, are reciting mere instructions to apply language such that it amounts to no more than mere instructions to apply the exceptions. In addition, the additional element of “store …” amounts to no more than insignificant extra-solution activity for storing data. Therefore, the additional elements do not integrate the abstract ideas into a practical application.
Step 2B Analysis: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, all of the additional elements are “mere instructions to apply an exception” (I.e. the additional elements describe a generic processor, accelerators, computing circuit, SRAM, model, and generic training of the model for applying the abstract ideas) or insignificant extra-solution activity (i.e. storing/retrieving data). Furthermore, the “store …” and “access the SRAM …” limitations are insignificant extra-solution activity that is well-understood, routine, and conventional according to MPEP 2106.05(d) (“The courts have recognized the following computer functions as well‐understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity… iv. Storing and retrieving information in memory). Mere instructions to apply an exception cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 8,
Claim 8 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 Analysis: Claim 8 is directed to an electronic system, which is directed to a machine, one of the statutory categories.
Step 2A Prong One Analysis: The limitation:
“generate the at least one probability value respectively corresponding to the at least one class within the corresponding category upon receiving of the data, for inferring which of the at least one class the received data falls within”
As drafted, under their broadest reasonable interpretations, cover mental processes (concepts performed in the human mind (including an observation, evaluation, judgement, opinion but for the recitation of mere instructions to apply language (See MPEP 2106.05(f)) and insignificant extra-solution activity (See MPEP 2106.05(g)). The above limitations in the context of this claim encompass generating a probability value corresponding to the class to infer which class the data falls within (corresponds to evaluation and judgement; in particular, a human, with the assistance of pen and paper, can generate a probability value for inferring which class the data falls within).
Step 2A Prong Two Analysis: The judicial exceptions are not integrated into a practical application. In particular, the claim recites additional elements that are mere instructions to apply (See MPEP 2106.05(f)) or insignificant extra-solution activity (See MPEP 2106.05(g)). The limitations:
“each accelerator”
As drafted, are additional elements that amount to no more than mere instructions to apply the exception for the abstract ideas. See MPEP 2106.05(f). In addition, the recitation of additional elements in claim 6 of a generic processor, accelerators, computing circuit, SRAM, model, and generic training of the model, as drafted, are reciting mere instructions to apply language such that it amounts to no more than mere instructions to apply the exceptions. In addition, the additional element of “store …” amounts to no more than insignificant extra-solution activity for storing data. Therefore, the additional elements do not integrate the abstract ideas into a practical application.
Step 2B Analysis: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, all of the additional elements are “mere instructions to apply an exception” (I.e. the additional elements describe a generic processor, accelerators, computing circuit, SRAM, model, and generic training of the model for applying the abstract ideas) or insignificant extra-solution activity (i.e. storing data). Furthermore, the “store …” limitation is insignificant extra-solution activity that is well-understood, routine, and conventional according to MPEP 2106.05(d) (“The courts have recognized the following computer functions as well‐understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity… iv. Storing and retrieving information in memory). Mere instructions to apply an exception cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 9,
Claim 9 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 Analysis: Claim 9 is directed to an electronic system, which is directed to a machine, one of the statutory categories.
Step 2A Prong One Analysis: The limitation:
“wherein each classification result further comprises a category probability value”
As drafted, is part of the abstract idea of claim 6 of generating a classification result. The limitation of claim 9 further limits the limitation of claim 6 by further defining what the classification result comprises. The above limitation in the context of this claim encompasses generating a classification result comprising a probability value and a category probability value to infer if data is part of the corresponding category (corresponds to evaluation and judgement; in particular, a human, with the assistance of pen and paper, can generate a classification comprising a probability value and a category probability value to determine if data falls within the corresponding category). The limitations:
“generate the category probability value upon receiving of the data, for inferring whether the data falls within the category”
As drafted, under their broadest reasonable interpretations, cover mental processes (concepts performed in the human mind (including an observation, evaluation, judgement, opinion but for the recitation of mere instructions to apply language (See MPEP 2106.05(f)) and insignificant extra-solution activity (See MPEP 2106.05(g)). The above limitations in the context of this claim encompass generating a category probability value to infer whether the data falls within the category (corresponds to evaluation and judgement; in particular, a human, with the assistance of pen and paper, can generate a category probability value for inferring if the data falls within the category).
Step 2A Prong Two Analysis: The judicial exceptions are not integrated into a practical application. In particular, the claim recites additional elements that are mere instructions to apply (See MPEP 2106.05(f)) or insignificant extra-solution activity (See MPEP 2106.05(g)). The limitations:
“each accelerator”
As drafted, are additional elements that amount to no more than mere instructions to apply the exception for the abstract ideas. See MPEP 2106.05(f). In addition, the recitation of additional elements in claim 8 of a generic processor, accelerators, computing circuit, SRAM, model, and generic training of the model, as drafted, are reciting mere instructions to apply language such that it amounts to no more than mere instructions to apply the exceptions. In addition, the additional element of “store …” amounts to no more than insignificant extra-solution activity for storing data. Therefore, the additional elements do not integrate the abstract ideas into a practical application.
Step 2B Analysis: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, all of the additional elements are “mere instructions to apply an exception” (I.e. the additional elements describe a generic processor, accelerators, computing circuit, SRAM, model, and generic training of the model for applying the abstract ideas) or insignificant extra-solution activity (i.e. storing data). Furthermore, the “store …” limitation is insignificant extra-solution activity that is well-understood, routine, and conventional according to MPEP 2106.05(d) (“The courts have recognized the following computer functions as well‐understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity… iv. Storing and retrieving information in memory). Mere instructions to apply an exception cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 10,
Claim 10 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 Analysis: Claim 10 is directed to an electronic system, which is directed to a machine, one of the statutory categories.
Step 2A Prong One Analysis: The limitations:
“wherein a summation of the category probability value and the at least one probability value of each classification result equals to 1”
As drafted, is part of the abstract ideas of claim 9 of generating a probability value and generating a category probability value. The limitation of claim 10 further limits the limitations of claim 9 by further defining the probability values. The above limitation in the context of this claim encompasses generating a probability value and generating a category probability value, wherein the summation of the probability values is equal to 1 (corresponds to evaluation and judgement; in particular, a human, with the assistance of pen and paper, can generate a category probability value and generate at least one probability value, with the summation of the probability values being 1).
Step 2A Prong Two Analysis: The judicial exceptions are not integrated into a practical application. In particular, the claim recites additional elements that are mere instructions to apply (See MPEP 2106.05(f)) or insignificant extra-solution activity (See MPEP 2106.05(g)). The recitation of additional elements in claim 9 of a generic processor, accelerators, computing circuit, SRAM, model, and generic training of the model, as drafted, are reciting mere instructions to apply language such that it amounts to no more than mere instructions to apply the exceptions. In addition, the additional element of “store …” amounts to no more than insignificant extra-solution activity for storing data. Therefore, the additional elements do not integrate the abstract ideas into a practical application.
Step 2B Analysis: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, all of the additional elements are “mere instructions to apply an exception” (I.e. the additional elements describe a generic processor, accelerators, computing circuit, SRAM, model, and generic training of the model for applying the abstract ideas) or insignificant extra-solution activity (i.e. storing data). Furthermore, the “store …” limitation is insignificant extra-solution activity that is well-understood, routine, and conventional according to MPEP 2106.05(d) (“The courts have recognized the following computer functions as well‐understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity… iv. Storing and retrieving information in memory). Mere instructions to apply an exception cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 11,
Claim 11 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 Analysis: Claim 11 is directed to an electronic system, which is directed to a machine, one of the statutory categories.
Step 2A Prong One Analysis: The limitations:
“upon receiving of the data, examining the category probability values generated by the plurality of accelerators to determine a selected category from the plurality of categories”
“examining the at least one probability value corresponding to the selected category to determine which class the data falls within”
As drafted, under their broadest reasonable interpretations, cover mental processes (concepts performed in the human mind (including an observation, evaluation, judgement, opinion)) but for the recitation of mere instructions to apply language (See MPEP 2106.05(f)) and insignificant extra-solution activity (See MPEP 2106.05(g)). The above limitations in the context of this claim encompass examining the category probability values to determine a selected category (corresponds to evaluation and judgement; in particular, a human, with the assistance of pen and paper, can determine a selected category of the plurality of categories by examining the category probability values); and examining the probability value corresponding to the selected category to determine which class the data falls within (corresponds to evaluation and judgement; in particular, a human, with the assistance of pen and paper, can determine which class the data falls within by examining the probability value corresponding to the selected category).
Step 2A Prong Two Analysis: The judicial exceptions are not integrated into a practical application. In particular, the claim recites additional elements that are mere instructions to apply (See MPEP 2106.05(f)) or insignificant extra-solution activity (See MPEP 2106.05(g)). The limitations:
“the processor”
As drafted, are additional elements that amount to no more than mere instructions to apply the exception for the abstract ideas. See MPEP 2106.05(f). In addition, the recitation of additional elements in claim 9 of a generic processor, accelerators, computing circuit, SRAM, model, and generic training of the model, as drafted, are reciting mere instructions to apply language such that it amounts to no more than mere instructions to apply the exceptions. In addition, the additional element of “store …” amounts to no more than insignificant extra-solution activity for storing data. Therefore, the additional elements do not integrate the abstract ideas into a practical application.
Step 2B Analysis: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, all of the additional elements are “mere instructions to apply an exception” (I.e. the additional elements describe a generic processor, accelerators, computing circuit, SRAM, model, and generic training of the model for applying the abstract ideas) or insignificant extra-solution activity (i.e. storing data). Furthermore, the “store …” limitation is insignificant extra-solution activity that is well-understood, routine, and conventional according to MPEP 2106.05(d) (“The courts have recognized the following computer functions as well‐understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity… iv. Storing and retrieving information in memory). Mere instructions to apply an exception cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 12,
Claim 12 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 Analysis: Claim 12 is directed to an electronic system, which is directed to a machine, one of the statutory categories.
Step 2A Prong One Analysis: The limitations:
“upon receiving of the data, … generating a plurality of category probability values respectively corresponding to the plurality of categories to infer which category the data falls within”
As drafted, under their broadest reasonable interpretations, cover mental processes (concepts performed in the human mind (including an observation, evaluation, judgement, opinion but for the recitation of mere instructions to apply language (See MPEP 2106.05(f)) and insignificant extra-solution activity (See MPEP 2106.05(g)). The above limitations in the context of this claim encompass generating category probability values corresponding to the categories to infer which category the data falls within (corresponds to evaluation and judgement; in particular, a human, with the assistance of pen and paper, can infer which category the data falls within by generating category probability values corresponding to the categories).
Step 2A Prong Two Analysis: The judicial exceptions are not integrated into a practical application. In particular, the claim recites additional elements that are mere instructions to apply (See MPEP 2106.05(f)) or insignificant extra-solution activity (See MPEP 2106.05(g)). The limitations:
“a category accelerator of the plurality of accelerators”
“the category accelerator”
“executing the category model”
As drafted, are additional elements that amount to no more than mere instructions to apply the exception for the abstract ideas. See MPEP 2106.05(f). The limitations:
“store a category model”
As drafted, are additional elements that correspond to insignificant extra-solution activity. In particular, the additional elements are merely directed towards mere data gathering. See MPEP 2106.05(g). In addition, the recitation of additional elements in claim 9 of a generic processor, accelerators, computing circuit, SRAM, model, and generic training of the model, as drafted, are reciting mere instructions to apply language such that it amounts to no more than mere instructions to apply the exceptions. In addition, the additional element of “store …” amounts to no more than insignificant extra-solution activity for storing data. Therefore, the additional elements do not integrate the abstract ideas into a practical application.
Step 2B Analysis: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, all of the additional elements are “mere instructions to apply an exception” (I.e. the additional elements describe a generic processor, accelerators, computing circuit, SRAM, model, and generic training of the model for applying the abstract ideas) or insignificant extra-solution activity (i.e. storing data). Furthermore, the “store …” limitations are insignificant extra-solution activity that is well-understood, routine, and conventional according to MPEP 2106.05(d) (“The courts have recognized the following computer functions as well‐understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity… iv. Storing and retrieving information in memory). Mere instructions to apply an exception cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 13,
Claim 13 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 Analysis: Claim 13 is directed to an electronic system, which is directed to a machine, one of the statutory categories.
Step 2A Prong One Analysis: The limitations:
“wherein after the category probability values are generated, … determine a selected category from the plurality of categories according to the category probability values”
As drafted, under their broadest reasonable interpretations, cover mental processes (concepts performed in the human mind (including an observation, evaluation, judgement, opinion)) but for the recitation of mere instructions to apply language (See MPEP 2106.05(f)) and insignificant extra-solution activity (See MPEP 2106.05(g)). The above limitations in the context of this claim encompass after generating the category probability values, determining a selected category according to the category probability values (corresponds to evaluation and judgement; in particular, a human, with the assistance of pen and paper, can determine a selected category based on the category probability values).
Step 2A Prong Two Analysis: The judicial exceptions are not integrated into a practical application. In particular, the claim recites additional elements that are mere instructions to apply (See MPEP 2106.05(f)) or insignificant extra-solution activity (See MPEP 2106.05(g)). The limitations:
“the processor”
As drafted, are additional elements that amount to no more than mere instructions to apply the exception for the abstract ideas. See MPEP 2106.05(f). In addition, the recitation of additional elements in claim 12 of a generic processor, accelerators, computing circuit, SRAM, model, and generic training of the model, as drafted, are reciting mere instructions to apply language such that it amounts to no more than mere instructions to apply the exceptions. In addition, the additional elements of “store …” amount to no more than insignificant extra-solution activity for storing data. Therefore, the additional elements do not integrate the abstract ideas into a practical application.
Step 2B Analysis: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, all of the additional elements are “mere instructions to apply an exception” (I.e. the additional elements describe a generic processor, accelerators, computing circuit, SRAM, model, and generic training of the model for applying the abstract ideas) or insignificant extra-solution activity (i.e. storing data). Furthermore, the “store …” limitations are insignificant extra-solution activity that is well-understood, routine, and conventional according to MPEP 2106.05(d) (“The courts have recognized the following computer functions as well‐understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity… iv. Storing and retrieving information in memory). Mere instructions to apply an exception cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 14,
Claim 14 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 Analysis: Claim 14 is directed to an electronic system, which is directed to a machine, one of the statutory categories.
Step 2A Prong One Analysis: The limitations:
“wherein after the selected category is determined, … generate at least one probability value respectively corresponding to at least one class within the selected category for inferring which class of the selected category the data falls within”
As drafted, under their broadest reasonable interpretations, cover mental processes (concepts performed in the human mind (including an observation, evaluation, judgement, opinion)) but for the recitation of mere instructions to apply language (See MPEP 2106.05(f)) and insignificant extra-solution activity (See MPEP 2106.05(g)). The above limitations in the context of this claim encompass after determining the selected category, generating a probability value corresponding to a class within the selected category for inferring which class of the selected category the data falls within (corresponds to evaluation and judgement; in particular, a human, with the assistance of pen and paper, can infer which class of the selected category the data falls within by generating a probability value corresponding to each class within the selected category).
Step 2A Prong Two Analysis: The judicial exceptions are not integrated into a practical application. In particular, the claim recites additional elements that are mere instructions to apply (See MPEP 2106.05(f)) or insignificant extra-solution activity (See MPEP 2106.05(g)). The limitations:
“the model corresponding to the selected category is configured to receive the data”
As drafted, are additional elements that correspond to insignificant extra-solution activity. In particular, the additional elements are merely directed towards mere data gathering. See MPEP 2106.05(g). In addition, the recitation of additional elements in claim 13 of a generic processor, accelerators, computing circuit, SRAM, model, and generic training of the model, as drafted, are reciting mere instructions to apply language such that it amounts to no more than mere instructions to apply the exceptions. In addition, the additional elements of “store …” amount to no more than insignificant extra-solution activity for storing data. Therefore, the additional elements do not integrate the abstract ideas into a practical application.
Step 2B Analysis: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, all of the additional elements are “mere instructions to apply an exception” (I.e. the additional elements describe a generic processor, accelerators, computing circuit, SRAM, model, and generic training of the model for applying the abstract ideas) or insignificant extra-solution activity (i.e. storing/receiving data). Furthermore, the “store …” and “… receive …” limitations are insignificant extra-solution activity that is well-understood, routine, and conventional according to MPEP 2106.05(d) (“The courts have recognized the following computer functions as well‐understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity… i. Receiving or transmitting data over a network … iv. Storing and retrieving information in memory). Mere instructions to apply an exception cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 15,
Claim 15 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 Analysis: Claim 15 is directed to an operating method, which is directed to a process, one of the statutory categories.
Step 2A Prong One Analysis: The limitations:
“upon receiving of a data, … generating a classification result comprising at least one probability value to infer whether the data falls within the corresponding category”
“establishing the plurality of categories, …, according to feature of each of a plurality of classes, wherein each of the plurality of categories includes at least one of the plurality of classes with same features”
As drafted, under their broadest reasonable interpretations, cover mental processes (concepts performed in the human mind (including an observation, evaluation, judgement, opinion)) but for the recitation of mere instructions to apply language (See MPEP 2106.05(f)) and insignificant extra-solution activity (See MPEP 2106.05(g)). The above limitations in the context of this claim encompass generating a classification result to infer if data is part of the corresponding category (corresponds to evaluation and judgement; in particular, a human, with the assistance of pen and paper, can generate a classification to determine if data falls within the corresponding category); and establishing categories according to a feature of each of a plurality of classes, the categories including classes with same features (corresponds to evaluation and judgement; in particular, a human, with the assistance of pen and paper, can establish categories of classes with same features).
Step 2A Prong Two Analysis: The judicial exceptions are not integrated into a practical application. In particular, the claim recites additional elements that are mere instructions to apply (See MPEP 2106.05(f)) or insignificant extra-solution activity (See MPEP 2106.05(g)). The limitations:
“an electronic system”
“providing a plurality of accelerators in the electronic system”
“each accelerator”
“the model is trained using a training dataset labeled with the at least one class”
“respectively forming each of the plurality of accelerators by a computing circuit and a on-chip static random access memory (SRAM) on each of a plurality of chips”
“making the computing circuit to execute, by each accelerator, the model”
“by the processor”
As drafted, are additional elements that amount to no more than mere instructions to apply the exception for the abstract ideas. See MPEP 2106.05(f). The limitations:
“store a model corresponding to one of a plurality of categories with at least one class having a same feature and being categorized within the category”
As drafted, are additional elements that correspond to insignificant extra-solution activity. In particular, the additional elements are merely directed towards mere data gathering. See MPEP 2106.05(g). Therefore, the additional elements do not integrate the abstract ideas into a practical application.
Step 2B Analysis: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, all of the additional elements are “mere instructions to apply an exception” (I.e. the additional elements describe a generic electronic system, accelerators, processor, computing circuit, SRAM, model, and generic training of the model for applying the abstract ideas) or insignificant extra-solution activity (i.e. storing data). Furthermore, the “store …” limitation is insignificant extra-solution activity that is well-understood, routine, and conventional according to MPEP 2106.05(d) (“The courts have recognized the following computer functions as well‐understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity… iv. Storing and retrieving information in memory). Mere instructions to apply an exception cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 16,
Claim 16 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 Analysis: Claim 16 is directed to an operating method, which is directed to a process, one of the statutory categories.
Step 2A Prong One Analysis: Please see the analysis of claim 15. The limitation of claim 16 is only an additional element to the abstract ideas of claim 15.
Step 2A Prong Two Analysis: The judicial exceptions are not integrated into a practical application. In particular, the claim recites additional elements that are mere instructions to apply (See MPEP 2106.05(f)) or insignificant extra-solution activity (See MPEP 2106.05(g)). The limitations:
“providing the SRAM”
“the computing circuit in each accelerator, coupled to the SRAM”
As drafted, are additional elements that amount to no more than mere instructions to apply the exception for the abstract ideas. See MPEP 2106.05(f). The limitations:
“store the corresponding model”
“access the SRAM to generate the classification result upon receiving of the data”
As drafted, are additional elements that correspond to insignificant extra-solution activity. In particular, the additional elements are merely directed towards mere data gathering. See MPEP 2106.05(g). In addition, the recitation of additional elements in claim 15 of a generic electronic system, accelerators, processor, computing circuit, SRAM, model, and generic training of the model, as drafted, are reciting mere instructions to apply language such that it amounts to no more than mere instructions to apply the exceptions. In addition, the additional element of “store …” amounts to no more than insignificant extra-solution activity for storing data. Therefore, the additional elements do not integrate the abstract ideas into a practical application.
Step 2B Analysis: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, all of the additional elements are “mere instructions to apply an exception” (I.e. the additional elements describe a generic electronic system, accelerators, processor, computing circuit, SRAM, model, and generic training of the model for applying the abstract ideas) or insignificant extra-solution activity (i.e. storing/retrieving data). Furthermore, the “store …” and “access the SRAM …” limitations are insignificant extra-solution activity that is well-understood, routine, and conventional according to MPEP 2106.05(d) (“The courts have recognized the following computer functions as well‐understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity… iv. Storing and retrieving information in memory). Mere instructions to apply an exception cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 17,
Claim 17 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 Analysis: Claim 17 is directed to an operating method, which is directed to a process, one of the statutory categories.
Step 2A Prong One Analysis: The limitation:
“generating … the at least one probability value respectively corresponding to the at least one class falls within the corresponding category upon receiving of the data, for inferring which of the at least one class the received data falls within”
As drafted, under their broadest reasonable interpretations, cover mental processes (concepts performed in the human mind (including an observation, evaluation, judgement, opinion but for the recitation of mere instructions to apply language (See MPEP 2106.05(f)) and insignificant extra-solution activity (See MPEP 2106.05(g)). The above limitations in the context of this claim encompass generating a probability value corresponding to the class to infer which class the data falls within (corresponds to evaluation and judgement; in particular, a human, with the assistance of pen and paper, can generate a probability value for inferring which class the data falls within).
Step 2A Prong Two Analysis: The judicial exceptions are not integrated into a practical application. In particular, the claim recites additional elements that are mere instructions to apply (See MPEP 2106.05(f)) or insignificant extra-solution activity (See MPEP 2106.05(g)). The limitations:
“by each accelerator”
As drafted, are additional elements that amount to no more than mere instructions to apply the exception for the abstract ideas. See MPEP 2106.05(f). In addition, the recitation of additional elements in claim 15 of a generic electronic system, accelerators, processor, computing circuit, SRAM, model, and generic training of the model, as drafted, are reciting mere instructions to apply language such that it amounts to no more than mere instructions to apply the exceptions. In addition, the additional element of “store …” amounts to no more than insignificant extra-solution activity for storing data. Therefore, the additional elements do not integrate the abstract ideas into a practical application.
Step 2B Analysis: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, all of the additional elements are “mere instructions to apply an exception” (I.e. the additional elements describe a generic electronic system, accelerators, processor, computing circuit, SRAM, model, and generic training of the model for applying the abstract ideas) or insignificant extra-solution activity (i.e. storing data). Furthermore, the “store …” limitation is insignificant extra-solution activity that is well-understood, routine, and conventional according to MPEP 2106.05(d) (“The courts have recognized the following computer functions as well‐understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity… iv. Storing and retrieving information in memory). Mere instructions to apply an exception cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 18,
Claim 18 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 Analysis: Claim 18 is directed to an operating method, which is directed to a process, one of the statutory categories.
Step 2A Prong One Analysis: The limitation:
“wherein each classification result further comprises a category probability value”
As drafted, is part of the abstract idea of claim 15 of generating a classification result. The limitation of claim 18 further limits the limitation of claim 15 by further defining what the classification result comprises. The above limitation in the context of this claim encompasses generating a classification result comprising a probability value and a category probability value to infer if data is part of the corresponding category (corresponds to evaluation and judgement; in particular, a human, with the assistance of pen and paper, can generate a classification comprising a probability value and a category probability value to determine if data falls within the corresponding category). The limitations:
“generating … the category probability value upon receiving of the data, for inferring whether the data falls within the category”
As drafted, under their broadest reasonable interpretations, cover mental processes (concepts performed in the human mind (including an observation, evaluation, judgement, opinion but for the recitation of mere instructions to apply language (See MPEP 2106.05(f)) and insignificant extra-solution activity (See MPEP 2106.05(g)). The above limitations in the context of this claim encompass generating a category probability value to infer whether the data falls within the category (corresponds to evaluation and judgement; in particular, a human, with the assistance of pen and paper, can generate a category probability value for inferring if the data falls within the category).
Step 2A Prong Two Analysis: The judicial exceptions are not integrated into a practical application. In particular, the claim recites additional elements that are mere instructions to apply (See MPEP 2106.05(f)) or insignificant extra-solution activity (See MPEP 2106.05(g)). The limitations:
“by each accelerator”
As drafted, are additional elements that amount to no more than mere instructions to apply the exception for the abstract ideas. See MPEP 2106.05(f). In addition, the recitation of additional elements in claim 17 of a generic electronic system, accelerators, processor, computing circuit, SRAM, model, and generic training of the model, as drafted, are reciting mere instructions to apply language such that it amounts to no more than mere instructions to apply the exceptions. In addition, the additional element of “store …” amounts to no more than insignificant extra-solution activity for storing data. Therefore, the additional elements do not integrate the abstract ideas into a practical application.
Step 2B Analysis: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, all of the additional elements are “mere instructions to apply an exception” (I.e. the additional elements describe a generic electronic system, accelerators, processor, computing circuit, SRAM, model, and generic training of the model for applying the abstract ideas) or insignificant extra-solution activity (i.e. storing data). Furthermore, the “store …” limitation is insignificant extra-solution activity that is well-understood, routine, and conventional according to MPEP 2106.05(d) (“The courts have recognized the following computer functions as well‐understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity… iv. Storing and retrieving information in memory). Mere instructions to apply an exception cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 19,
Claim 19 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 Analysis: Claim 19 is directed to an operating method, which is directed to a process, one of the statutory categories.
Step 2A Prong One Analysis: The limitations:
“wherein a summation of the category probability value and 25 the at least one probability value of each classification result equals to 1”
As drafted, is part of the abstract ideas of claim 18 of generating a probability value and generating a category probability value. The limitation of claim 19 further limits the limitations of claim 18 by further defining the probability values. The above limitation in the context of this claim encompasses generating a probability value and generating a category probability value, wherein the summation of the probability values is equal to 1 (corresponds to evaluation and judgement; in particular, a human, with the assistance of pen and paper, can generate a category probability value and generate at least one probability value, with the summation of the probability values being 1).
Step 2A Prong Two Analysis: The judicial exceptions are not integrated into a practical application. In particular, the claim recites additional elements that are mere instructions to apply (See MPEP 2106.05(f)) or insignificant extra-solution activity (See MPEP 2106.05(g)). The recitation of additional elements in claim 18 of a generic electronic system, accelerators, processor, computing circuit, SRAM, model, and generic training of the model, as drafted, are reciting mere instructions to apply language such that it amounts to no more than mere instructions to apply the exceptions. In addition, the additional element of “store …” amounts to no more than insignificant extra-solution activity for storing data. Therefore, the additional elements do not integrate the abstract ideas into a practical application.
Step 2B Analysis: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, all of the additional elements are “mere instructions to apply an exception” (I.e. the additional elements describe a generic electronic system, accelerators, processor, computing circuit, SRAM, model, and generic training of the model for applying the abstract ideas) or insignificant extra-solution activity (i.e. storing data). Furthermore, the “store …” limitation is insignificant extra-solution activity that is well-understood, routine, and conventional according to MPEP 2106.05(d) (“The courts have recognized the following computer functions as well‐understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity… iv. Storing and retrieving information in memory). Mere instructions to apply an exception cannot provide an inventive concept. The claim is not patent eligible.
Regarding Claim 20,
Claim 20 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 Analysis: Claim 20 is directed to an operating method, which is directed to a process, one of the statutory categories.
Step 2A Prong One Analysis: The limitations:
“upon receiving of the data, examining … the category probability values generated by the plurality of accelerators to determine a selected category which the data falls within”
“examining … the at least one class probability value corresponding to the selected category to determine which class the data falls within”
As drafted, under their broadest reasonable interpretations, cover mental processes (concepts performed in the human mind (including an observation, evaluation, judgement, opinion)) but for the recitation of mere instructions to apply language (See MPEP 2106.05(f)) and insignificant extra-solution activity (See MPEP 2106.05(g)). The above limitations in the context of this claim encompass examining the category probability values to determine a selected category (corresponds to evaluation and judgement; in particular, a human, with the assistance of pen and paper, can determine a selected category of the plurality of categories by examining the category probability values); and examining a class probability value corresponding to the selected category to determine which class the data falls within (corresponds to evaluation and judgement; in particular, a human, with the assistance of pen and paper, can determine which class the data falls within by examining the class probability value corresponding to the selected category).
Step 2A Prong Two Analysis: The judicial exceptions are not integrated into a practical application. In particular, the claim recites additional elements that are mere instructions to apply (See MPEP 2106.05(f)) or insignificant extra-solution activity (See MPEP 2106.05(g)). The limitations:
“by a processor”
“by the processor”
As drafted, are additional elements that amount to no more than mere instructions to apply the exception for the abstract ideas. See MPEP 2106.05(f). In addition, the recitation of additional elements in claim 18 of a generic electronic system, accelerators, processor, computing circuit, SRAM, model, and generic training of the model, as drafted, are reciting mere instructions to apply language such that it amounts to no more than mere instructions to apply the exceptions. In addition, the additional element of “store …” amounts to no more than insignificant extra-solution activity for storing data. Therefore, the additional elements do not integrate the abstract ideas into a practical application.
Step 2B Analysis: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, all of the additional elements are “mere instructions to apply an exception” (I.e. the additional elements describe a generic electronic system, accelerators, processor, computing circuit, SRAM, model, and generic training of the model for applying the abstract ideas) or insignificant extra-solution activity (i.e. storing data). Furthermore, the “store …” limitation is insignificant extra-solution activity that is well-understood, routine, and conventional according to MPEP 2106.05(d) (“The courts have recognized the following computer functions as well‐understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity… iv. Storing and retrieving information in memory). Mere instructions to apply an exception cannot provide an inventive concept. The claim is not patent eligible.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 1-5 are rejected under 35 U.S.C. 103 as being unpatentable over Rossi et al. (US 2020/0310761 A1) in view of Li et al. (US 2014/0101119 A1) in view of Schiff et al. (US 2009/0172730 A1) and further in view of Marrero et al. (US 2022/0108134 A1).
Regarding Claim 1,
Rossi et al. teaches a training method of a model applied to an accelerator ([0115]-[0116]: "FIG. 1 presents a schematic overview of a classification task, in which an image is analyzed by a CNN to classify a photograph. After deployment, the classifier autonomously assigns the correct label to a test subject (in this case, to images). Alternatively, it could assign to each label a certain probability of membership to a class. The actual realization depends, of course, on the specific application and the classification method is chosen at design time. An example is depicted in FIG. 2, where the classifier processes the image of a cat. … There are various methods to realize a classifier with regard to how it performs classification. In a data-driven approach, the classifier undergoes a training process before its deployment. The training includes letting the classifier process a set of inputs whose ground truth is known a priori by the designer, the classification criteria being determined by a certain set of parameters" teaches a training process (method) of a classifier model for a classification task. [0109]: "In an exemplary framework for accelerating neural networks, data are transferred within the framework in a streamed fashion, and multiple multiply-and-accumulate (MAC) units are clustered together to exploit CNNs' parameter sharing property. All MACs in a cluster work in parallel, allowing to compute multiple kernels at the same to increase the throughput. A number of these clusters together form a convolution accelerator (CA)" teaches that the neural network framework (model) is applied to an accelerator), the training method comprising:
receiving a training dataset having a plurality of training data by a processor, wherein each training data is labeled to one of a plurality of classes ([0115]-[0116]: "FIG. 1 presents a schematic overview of a classification task, in which an image is analyzed by a CNN to classify a photograph. After deployment, the classifier autonomously assigns the correct label to a test subject (in this case, to images). Alternatively, it could assign to each label a certain probability of membership to a class. The actual realization depends, of course, on the specific application and the classification method is chosen at design time. An example is depicted in FIG. 2, where the classifier processes the image of a cat. … There are various methods to realize a classifier with regard to how it performs classification. In a data-driven approach, the classifier undergoes a training process before its deployment. The training includes letting the classifier process a set of inputs whose ground truth is known a priori by the designer, the classification criteria being determined by a certain set of parameters" teaches that a set of inputs for a training process (training dataset) is received, the inputs being labeled with a ground truth classification);
training a first model with the training dataset labeled with the plurality of classes (Fig. 1; Fig. 2; [0115]-[0116]: "FIG. 1 presents a schematic overview of a classification task, in which an image is analyzed by a CNN to classify a photograph. After deployment, the classifier autonomously assigns the correct label to a test subject (in this case, to images). Alternatively, it could assign to each label a certain probability of membership to a class. The actual realization depends, of course, on the specific application and the classification method is chosen at design time. An example is depicted in FIG. 2, where the classifier processes the image of a cat. … There are various methods to realize a classifier with regard to how it performs classification. In a data-driven approach, the classifier undergoes a training process before its deployment. The training includes letting the classifier process a set of inputs whose ground truth is known a priori by the designer, the classification criteria being determined by a certain set of parameters" teaches that a set of inputs (training dataset) with known ground truths (labels for a plurality of classes) is used in the training process for a classifier model (first model) (i.e. the classifier model is trained with labeled input data)), and
implementing the first model on the accelerator by the computing circuit and the memory (Fig. 74; [0078]: "FIG. 74 depicts an embodiment of a processor-based device in which a convolution accelerator operates in conjunction with a system-on-chip and a coupled co-processor subsystem to provide an accelerated neural network in accordance with one or more techniques presented herein" teaches that the accelerator implements the neural network (first model)).
Rossi et al. does not appear to explicitly teach establishing a plurality of categories according to feature of each of a plurality of classes by the processor, wherein each of the plurality of categories includes at least one of the plurality of classes with same features; and using at least one first class within one of the plurality of categories for verification by the processor; disposing a computing circuit and a memory on a same chip to form the accelerator; and accessing the memory internally in the chip in order to execute the corresponding model by the computing circuit for generating the classification result upon receiving of the data.
However, Li et al. teaches establishing a plurality of categories according to feature of each of a plurality of classes by the processor, wherein each of the plurality of categories includes at least one of the plurality of classes with same features (Fig. 4; Fig. 5; [0070]: "In FIG. 5, a query is evaluated 510 by a plurality of domain classifiers. The query is evaluated to determine a plurality of domain evaluation results. Category scores are generated 520 for meta-classifier categories based on the plurality of domain evaluation results. Preferably, the category scores are generated by meta-classifiers that are based on a non-linear classification model. In the embodiment shown in FIG. 5, at least one category score is greater than a threshold value. A meta-classifier category is then selected 530 based on the at least one category score that is greater than the threshold value. A domain is assigned 540 to the query based on the selected meta-classifier category" teaches that the meta-classifier generates category (class) scores based on domain evaluation results (features) and that the category (class) score that is greater than a threshold (at least one first class) is selected to assign (establish) a domain (category) having the category (class) score (i.e. a domain (category) is established based on selecting a category (class) score from the plurality of category (class) scores having a same domain evaluation result (feature)). [0050]-[0051]: "the initial features for the meta-classifiers are selected to be the output scores from the domain classifiers, plus various additional feature used by one or more of the domain classifiers ... In order to train the meta-classifiers, the labeled queries in a training set are first evaluated by the domain classifiers (which have already been trained) to generate domain evaluation scores. The evaluation scores are then aggregated for use by each meta-classifier. In this example, the meta-classifiers are designed to provide a probability of association between a query and a meta-classifier category. Because some domains do not have a corresponding meta-classifier, queries with a marginal association should not necessarily be associated with a meta-classifier category. As a result, a meta-classifier threshold level is set at 50%. If a meta-classifier provides an association probability of lower than 50%, then the query is determined to not be associated with the corresponding meta-classifier category. If at least one category score is greater than 50%, the query is assigned to one or more meta-classifier categories. The meta-classifier with the highest probability (or other category score) is used to assign the query to the corresponding category. This continues for each meta-classifier with a category score greater than 50%. Optionally, if two or more of the meta-classifiers have a mutually exclusive relationship, a category assignment of the query to a first meta-classifier category is a second mutually exclusive category has already been assigned (such as based on a higher category score)" teaches that the domain output results (domain evaluation results) from the domain classifiers are features for the meta-classifiers used to establish categories based on category (class) scores from the plurality of category (class) scores having a same domain evaluation result (feature))).
Rossi et al. is analogous to the claimed invention because it is directed to the implementation of a neural network accelerator.
Li et al. is analogous to the claimed invention because it is directed to neural network classification.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate establishing a plurality of categories according to feature of each of a plurality of classes by the processor, wherein each of the plurality of categories includes at least one of the plurality of classes with same features as taught by Li et al. to the disclosed invention of Rossi et al.
One of ordinary skill in the art would have been motivated to make this modification to provide "improved methods for assigning such queries to the correct category without requiring excessive additional resources" (Li et al. [0001]).
Rossi et al. in view of Li et al. does not appear to explicitly teach using at least one first class within one of the plurality of categories for verification by the processor; disposing a computing circuit and a memory on a same chip to form the accelerator; and accessing the memory internally in the chip in order to execute the corresponding model by the computing circuit for generating the classification result upon receiving of the data.
However, Schiff et al. teaches using at least one first class within one of the plurality of categories for verification by the processor (Fig. 6B; [0075]-[0076]: "In one embodiment, the machine learning module 210 generates predetermined models for classifying images based on a hierarchical method. The hierarchical method fuses multiple binary classifiers into a k-classifier. In general, in a binary classifier, the more likely of two classes can be determined and thus selected. Thus, for multi-class classification, fusing of multiple binary classifiers can be performed. … In one embodiment, the k-classifier is built from results of a generalization accuracy computation. The generalization accuracy computation is, in one embodiment, determined from performing machine learning on a learning data set and recording the accuracy on a verification data set. For example, the training data set can be split into a learning data set and a verification data set. Machine learning can then be performed on the learning data set to generate a model. The accuracy of the model can then be determined by applying the model to the verification data set. By performing a learning process and a verification process, a probability for each pair of categories can be generated" teaches that the model is trained and verified during the machine learning process by using the verification data set, which comprises accuracies/probabilities based on the classes (at least one first class) of the category (one of the plurality of categories)).
Rossi et al. is analogous to the claimed invention because it is directed to the implementation of a neural network accelerator.
Li et al. and Schiff et al. are analogous to the claimed invention because they are directed to neural network classification.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate using at least one first class within one of the plurality of categories for verification by the processor as taught by Schiff et al. to the disclosed invention of Rossi et al. in view of Li et al.
One of ordinary skill in the art would have been motivated to make this modification "in order to improve accuracy and speed of image classification" (Schiff et al. [0014]).
Rossi et al. in view of Li et al. and further in view of Schiff et al. does not appear to explicitly teach disposing a computing circuit and a memory on a same chip to form the accelerator; and accessing the memory internally in the chip in order to execute the corresponding model by the computing circuit for generating the classification result upon receiving of the data.
However, Marrero et al. teaches disposing a computing circuit and a memory on a same chip to form the accelerator (Fig. 7A; Fig. 8; [0080]-[0082]: "In at least one embodiment, inference and/or training logic 715 may include, without limitation, a code and/or data storage 705 to store backward and/or output weight and/or input/output data corresponding to neurons or layers of a neural network trained and/or used for inferencing in aspects of one or more embodiments. In at least one embodiment, code and/or data storage 705 stores weight parameters and/or input/output data of each layer of a neural network trained or used in conjunction with one or more embodiments during backward propagation of input/output data and/or weight parameters during training and/or inferencing using aspects of one or more embodiments. … In at least one embodiment, code and/or data storage 705 may be cache memory, DRAM, SRAM, non-volatile memory (e.g., Flash memory), or other storage … In at least one embodiment, inference and/or training logic 715 may include, without limitation, one or more arithmetic logic unit(s) (“ALU(s)”) 710, including integer and/or floating point units, to perform logical and/or mathematical operations based, at least in part on, or indicated by, training and/or inference code (e.g., graph code), a result of which may produce activations (e.g., output values from layers or neurons within a neural network) stored in an activation storage 720 that are functions of input/output and/or weight parameter data stored in code and/or data storage 701 and/or code and/or data storage 705" teaches that each node computing resource 816 (accelerator) comprises inference and/or training logic 715, which comprises a data storage 705 (memory), the data storage 705 being comprised of SRAM, and ALUs 710 (computing circuit) that performs operations of a neural network on a same chip); and
accessing the memory internally in the chip in order to execute the corresponding model by the computing circuit for generating the classification result upon receiving of the data (Fig. 7A; Fig. 8; [0080]-[0082]: "In at least one embodiment, inference and/or training logic 715 may include, without limitation, a code and/or data storage 705 to store backward and/or output weight and/or input/output data corresponding to neurons or layers of a neural network trained and/or used for inferencing in aspects of one or more embodiments. In at least one embodiment, code and/or data storage 705 stores weight parameters and/or input/output data of each layer of a neural network trained or used in conjunction with one or more embodiments during backward propagation of input/output data and/or weight parameters during training and/or inferencing using aspects of one or more embodiments. … In at least one embodiment, code and/or data storage 705 may be cache memory, DRAM, SRAM, non-volatile memory (e.g., Flash memory), or other storage … In at least one embodiment, inference and/or training logic 715 may include, without limitation, one or more arithmetic logic unit(s) (“ALU(s)”) 710, including integer and/or floating point units, to perform logical and/or mathematical operations based, at least in part on, or indicated by, training and/or inference code (e.g., graph code), a result of which may produce activations (e.g., output values from layers or neurons within a neural network) stored in an activation storage 720 that are functions of input/output and/or weight parameter data stored in code and/or data storage 701 and/or code and/or data storage 705" teaches that each node computing resource 816 (accelerator) comprises inference and/or training logic 715, which comprises a data storage 705 that stores information for a training neural network (corresponding model), the data storage 705 being comprised of SRAM, and ALUs 710 (computing circuit) that performs operations of a neural network, meaning that the memory of the accelerator is accessed to performs operations of a neural network using the computing circuit to generate outputs (executes corresponding model to generate classification result)).
Rossi et al. is analogous to the claimed invention because it is directed to the implementation of a neural network accelerator.
Li et al., Schiff et al., and Marrero et al. are analogous to the claimed invention because they are directed to neural network classification.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate disposing a computing circuit and a memory on a same chip to form the accelerator; and accessing the memory internally in the chip in order to execute the corresponding model by the computing circuit for generating the classification result upon receiving of the data as taught by Marrero et al. to the disclosed invention of Rossi et al. in view of Li et al. and further in view of Schiff et al.
One of ordinary skill in the art would have been motivated to make this modification to "eliminate need to transfer smaller units of data across processor's data bus to perform one or more operations one data element at a time" (Marrero et al. [0103]).
Regarding Claim 2,
Rossi et al. in view of Li et al. in view of Schiff et al. and further in view of Marrero et al. teaches the training method of claim 1.
In addition, Rossi et al. further teaches wherein upon receiving of each training data, the trained first model is configured to generate at least one first probability value, respectively corresponding to the at least one first class, for inferring percentages an object of the at least one class is shown in each training data (Fig. 1; [0115]-[0116]: "FIG. 1 presents a schematic overview of a classification task, in which an image is analyzed by a CNN to classify a photograph. After deployment, the classifier autonomously assigns the correct label to a test subject (in this case, to images). Alternatively, it could assign to each label a certain probability of membership to a class. The actual realization depends, of course, on the specific application and the classification method is chosen at design time. An example is depicted in FIG. 2, where the classifier processes the image of a cat. … There are various methods to realize a classifier with regard to how it performs classification. In a data-driven approach, the classifier undergoes a training process before its deployment. The training includes letting the classifier process a set of inputs whose ground truth is known a priori by the designer, the classification criteria being determined by a certain set of parameters" teaches that an input image (training data) can be classified into different categories such as car, person, or animal. Fig. 2; [0115]-[0116]: "FIG. 1 presents a schematic overview of a classification task, in which an image is analyzed by a CNN to classify a photograph. After deployment, the classifier autonomously assigns the correct label to a test subject (in this case, to images). Alternatively, it could assign to each label a certain probability of membership to a class. The actual realization depends, of course, on the specific application and the classification method is chosen at design time. An example is depicted in FIG. 2, where the classifier processes the image of a cat. … There are various methods to realize a classifier with regard to how it performs classification. In a data-driven approach, the classifier undergoes a training process before its deployment. The training includes letting the classifier process a set of inputs whose ground truth is known a priori by the designer, the classification criteria being determined by a certain set of parameters" teaches that probabilities for classes within a category (at least one first probability value) are generated for the input image (training data), with the probabilities being inferred percentages of the object in the image being of the at least one class (e.g. for an image in the animal category, the probability of a cat class is 82% and of a dog class is 15% are classes within the category)).
Regarding Claim 3,
Rossi et al. in view of Li et al. in view of Schiff et al. and further in view of Marrero et al. teaches the training method of claim 2.
In addition, Rossi et al. further teaches wherein upon receiving of each training data, the trained first model is further configured to generate a first category probability value, for inferring a percentage whether objects of the first category is shown in each training data (Fig. 1; [0115]-[0116]: "FIG. 1 presents a schematic overview of a classification task, in which an image is analyzed by a CNN to classify a photograph. After deployment, the classifier autonomously assigns the correct label to a test subject (in this case, to images). Alternatively, it could assign to each label a certain probability of membership to a class. The actual realization depends, of course, on the specific application and the classification method is chosen at design time. An example is depicted in FIG. 2, where the classifier processes the image of a cat. … There are various methods to realize a classifier with regard to how it performs classification. In a data-driven approach, the classifier undergoes a training process before its deployment. The training includes letting the classifier process a set of inputs whose ground truth is known a priori by the designer, the classification criteria being determined by a certain set of parameters" teaches that an input image (training data) can be classified into different categories such as car, person, or animal. Fig. 2; [0115]-[0116]: "FIG. 1 presents a schematic overview of a classification task, in which an image is analyzed by a CNN to classify a photograph. After deployment, the classifier autonomously assigns the correct label to a test subject (in this case, to images). Alternatively, it could assign to each label a certain probability of membership to a class. The actual realization depends, of course, on the specific application and the classification method is chosen at design time. An example is depicted in FIG. 2, where the classifier processes the image of a cat. … There are various methods to realize a classifier with regard to how it performs classification. In a data-driven approach, the classifier undergoes a training process before its deployment. The training includes letting the classifier process a set of inputs whose ground truth is known a priori by the designer, the classification criteria being determined by a certain set of parameters" teaches that probabilities for classes outside of a category (first category probability value) are generated for the input image (training data), with the probabilities being inferred percentages of the object in the image being outside of the category (e.g. for an image in the animal category, the probability of a hat class is 2% and of a mug class is 1% are classes outside the category, meaning that the percentage of the object not being in the first category is 3%)).
Regarding Claim 4,
Rossi et al. in view of Li et al. in view of Schiff et al. and further in view of Marrero et al. teaches the training method of claim 3.
In addition, Rossi et al. further teaches wherein a summation of the first category probability value and the at least one first probability value equals to 1 (Fig. 1; [0115]: "FIG. 1 presents a schematic overview of a classification task, in which an image is analyzed by a CNN to classify a photograph. After deployment, the classifier autonomously assigns the correct label to a test subject (in this case, to images). Alternatively, it could assign to each label a certain probability of membership to a class. The actual realization depends, of course, on the specific application and the classification method is chosen at design time. An example is depicted in FIG. 2, where the classifier processes the image of a cat" teaches that an image can be classified into different categories such as car, person, or animal. Fig. 2; [0115]: "FIG. 1 presents a schematic overview of a classification task, in which an image is analyzed by a CNN to classify a photograph. After deployment, the classifier autonomously assigns the correct label to a test subject (in this case, to images). Alternatively, it could assign to each label a certain probability of membership to a class. The actual realization depends, of course, on the specific application and the classification method is chosen at design time. An example is depicted in FIG. 2, where the classifier processes the image of a cat" teaches that the image can be assigned probabilities for classes within a category (at least one first probability value) and for classes outside of the category (category probability value), with the summation of the probabilities of each classification result (cat is 0.82 and dog is 0.15 as classes within the animal category) with the category probability value (hat is 0.02 and mug is 0.01 for classes outside of the animal category) is equal to 1).
Regarding Claim 5,
Rossi et al. in view of Li et al. in view of Schiff et al. and further in view of Marrero et al. teaches the training method of claim 1.
In addition, Schiff et al. further teaches wherein the step of training the first model with the training dataset and using the at least one first class falls within the first category for verification comprising: establishing a first category class by merging all classes that fall outside of the first category (Fig. 6B; [0075]-[0079]: "In one embodiment, the machine learning module 210 generates predetermined models for classifying images based on a hierarchical method. The hierarchical method fuses multiple binary classifiers into a k-classifier. In general, in a binary classifier, the more likely of two classes can be determined and thus selected. Thus, for multi-class classification, fusing of multiple binary classifiers can be performed. … In one embodiment, the k-classifier is built from results of a generalization accuracy computation. The generalization accuracy computation is, in one embodiment, determined from performing machine learning on a learning data set and recording the accuracy on a verification data set. For example, the training data set can be split into a learning data set and a verification data set. Machine learning can then be performed on the learning data set to generate a model. The accuracy of the model can then be determined by applying the model to the verification data set. By performing a learning process and a verification process, a probability for each pair of categories can be generated. … The probability for each pair of categories obtained from the combination of the learning and verification process can then, in one embodiment, be used to construct a `tree` to facilitate the multi-classification process. … For example, the tree construction can occur from the bottom-up or from the top-down. One example of the bottom-up scheme is to merge two categories with the least generalization accuracy and merging them to create a new category (sometimes also referred to as a `meta-category`). The creation of the new category also implicitly creates a new node with the two original categories as child-categories. Subsequently, in one embodiment, the machine learning process can be performed for the situation where the two categories are treated as one category. The new accuracies for the newly constructed model based on having merged categories are determined" teaches that the categories (classes) with the lowest generalization accuracy (e.g. fall outside the first category) are merged to create a new meta-category (first category class));
training the first model with the training dataset (Fig. 6B; [0079]-[0080]: "For example, the tree construction can occur from the bottom-up or from the top-down. One example of the bottom-up scheme is to merge two categories with the least generalization accuracy and merging them to create a new category (sometimes also referred to as a `meta-category`). The creation of the new category also implicitly creates a new node with the two original categories as child-categories. Subsequently, in one embodiment, the machine learning process can be performed for the situation where the two categories are treated as one category. The new accuracies for the newly constructed model based on having merged categories are determined. … This process can be repeated iteratively until there are two categories left among which, the better fitting one can be determined therefore yielding a categorization that occurs in an inverse direction" teaches that the merged category is used in the machine learning process (training with the training dataset) to generate new accuracies); and
verifying the first model by using the first category class and the at least one first class (Fig. 6B; [0076]-[0080]: "In one embodiment, the k-classifier is built from results of a generalization accuracy computation. The generalization accuracy computation is, in one embodiment, determined from performing machine learning on a learning data set and recording the accuracy on a verification data set. For example, the training data set can be split into a learning data set and a verification data set. Machine learning can then be performed on the learning data set to generate a model. The accuracy of the model can then be determined by applying the model to the verification data set. By performing a learning process and a verification process, a probability for each pair of categories can be generated. … The probability for each pair of categories obtained from the combination of the learning and verification process can then, in one embodiment, be used to construct a `tree` to facilitate the multi-classification process. … For example, the tree construction can occur from the bottom-up or from the top-down. One example of the bottom-up scheme is to merge two categories with the least generalization accuracy and merging them to create a new category (sometimes also referred to as a `meta-category`). The creation of the new category also implicitly creates a new node with the two original categories as child-categories. Subsequently, in one embodiment, the machine learning process can be performed for the situation where the two categories are treated as one category. The new accuracies for the newly constructed model based on having merged categories are determined … This process can be repeated iteratively until there are two categories left among which, the better fitting one can be determined therefore yielding a categorization that occurs in an inverse direction" teaches that the model is verified during the machine learning process by using the verification data set, which comprises accuracies/probabilities based on the merged category (first category class) and the remaining most likely category/class (at least one first class)).
Rossi et al. is analogous to the claimed invention because it is directed to the implementation of a neural network accelerator.
Li et al., Schiff et al., and Marrero et al. are analogous to the claimed invention because they are directed to neural network classification.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate wherein the step of training the first model with the training dataset and using the at least one first class falls within the first category for verification comprising: establishing a first category class by merging all classes that fall outside of the first category; training the first model with the training dataset; and verifying the first model by using the first category class and the at least one first class as taught by Schiff et al. to the disclosed invention of Rossi et al. in view of Li et al. and further in view of Marrero et al.
One of ordinary skill in the art would have been motivated to make this modification "in order to improve accuracy and speed of image classification" (Schiff et al. [0014]).
Claims 6-20 are rejected under 35 U.S.C. 103 as being unpatentable over Marrero et al. (US 2022/0108134 A1) in view of Li et al. (US 2014/0101119 A1) and further in view of Rossi et al. (US 2020/0310761 A1).
Regarding Claim 6,
Marrero et al. teaches an electronic system, comprising: a processor; and a plurality of accelerators, coupled to the processor (Fig. 7A; Fig. 8; [0097]-[0098]: "Inference and/or training logic 715 are used to perform inferencing and/or training operations associated with one or more embodiments. Details regarding inference and/or training logic 715 are provided below in conjunction with FIGS. 7A and/or 7B. In at least one embodiment, inference and/or training logic 715 may be used in system FIG. 8 for inferencing or predicting operations based, at least in part, on weight parameters calculated using neural network training operations, neural network functions and/or architectures, or neural network use cases described herein. … one or more neural networks, as may include a feature extraction network and a classifier network" teaches a system for performing inferencing or predicting operations using a neural network, including a classifier network. Fig. 8; [0088]: "In at least one embodiment, as shown in FIG. 8, data center infrastructure layer 810 may include a resource orchestrator 812, grouped computing resources 814, and node computing resources (“node C.R.s”) 816(1)-816(N), where “N” represents any whole, positive integer. In at least one embodiment, node C.R.s 816(1)-816(N) may include, but are not limited to, any number of central processing units (“CPUs”) or other processors (including accelerators, field programmable gate arrays (FPGAs), graphics processors, etc.), memory devices (e.g., dynamic read-only memory), storage devices (e.g., solid state or disk drives), network input/output (“NW I/O”) devices, network switches, virtual machines (“VMs”), power modules, and cooling modules, etc." teaches that the system comprises grouped computing resources 814 (processor), and node computing resources 816 (plurality of accelerators) for implementing the neural network),
wherein each accelerator comprises a computing circuit and a on-chip static random access memory (SRAM) (Fig. 7A; Fig. 8; [0080]-[0082]: "In at least one embodiment, inference and/or training logic 715 may include, without limitation, a code and/or data storage 705 to store backward and/or output weight and/or input/output data corresponding to neurons or layers of a neural network trained and/or used for inferencing in aspects of one or more embodiments. In at least one embodiment, code and/or data storage 705 stores weight parameters and/or input/output data of each layer of a neural network trained or used in conjunction with one or more embodiments during backward propagation of input/output data and/or weight parameters during training and/or inferencing using aspects of one or more embodiments. … In at least one embodiment, code and/or data storage 705 may be cache memory, DRAM, SRAM, non-volatile memory (e.g., Flash memory), or other storage … In at least one embodiment, inference and/or training logic 715 may include, without limitation, one or more arithmetic logic unit(s) (“ALU(s)”) 710, including integer and/or floating point units, to perform logical and/or mathematical operations based, at least in part on, or indicated by, training and/or inference code (e.g., graph code), a result of which may produce activations (e.g., output values from layers or neurons within a neural network) stored in an activation storage 720 that are functions of input/output and/or weight parameter data stored in code and/or data storage 701 and/or code and/or data storage 705" teaches that each node computing resource 816 (accelerator) comprises inference and/or training logic 715, which comprises a data storage 705 (memory), the data storage 705 being comprised of SRAM, and ALUs 710 (computing circuit) that performs operations of a neural network on a same chip), and
wherein the plurality of accelerators and corresponding SRAMs are respectively disposed on a plurality of chips (Fig. 7A; Fig. 8; [0080]-[0082]: "In at least one embodiment, inference and/or training logic 715 may include, without limitation, a code and/or data storage 705 to store backward and/or output weight and/or input/output data corresponding to neurons or layers of a neural network trained and/or used for inferencing in aspects of one or more embodiments. In at least one embodiment, code and/or data storage 705 stores weight parameters and/or input/output data of each layer of a neural network trained or used in conjunction with one or more embodiments during backward propagation of input/output data and/or weight parameters during training and/or inferencing using aspects of one or more embodiments. … In at least one embodiment, code and/or data storage 705 may be cache memory, DRAM, SRAM, non-volatile memory (e.g., Flash memory), or other storage … In at least one embodiment, inference and/or training logic 715 may include, without limitation, one or more arithmetic logic unit(s) (“ALU(s)”) 710, including integer and/or floating point units, to perform logical and/or mathematical operations based, at least in part on, or indicated by, training and/or inference code (e.g., graph code), a result of which may produce activations (e.g., output values from layers or neurons within a neural network) stored in an activation storage 720 that are functions of input/output and/or weight parameter data stored in code and/or data storage 701 and/or code and/or data storage 705" teaches that each node computing resource 816 (accelerator) comprises inference and/or training logic 715, which comprises a data storage 705 (memory), the data storage 705 being comprised of SRAM, and ALUs 710 (computing circuit) that performs operations of a neural network on a same chip).
Marrero et al. does not appear to explicitly teach each accelerator being configured to store a model corresponding to one of a plurality of categories with at least one class having a same feature and being categorized within the category, wherein the model is trained using a training dataset labeled with the at least one class, and each accelerator is configured to perform: upon receiving of a data, making the computing circuit to execute the model for generating a classification result comprising at least one probability value to infer whether the data falls within the corresponding category, and the plurality of accelerators are respectively corresponding to the plurality of categories, wherein the processor is configured to: establish the plurality of categories according to feature of each of a plurality of classes, wherein each of the plurality of categories includes at least one of the plurality of classes with same features.
However, Li et al. teaches each accelerator being configured to store a model corresponding to one of a plurality of categories with at least one class having a same feature and being categorized within the category (Fig. 4; Fig. 5; [0016]: "A first group of query classifiers can be used to evaluate a query relative to various subject matter domains. This initial evaluation by the domain classifiers provides some type of ranking, probability, or other score for a query relative to the domains. The evaluation output or results from the first group of domain classifiers can then be used by a second group of meta-classifiers. The meta-classifiers are associated with subject matter categories that may correspond to a domain, or that may correspond to a plurality of domains. Because the meta-classifiers are limited in scope, the meta-classifiers can be trained to use the output from the domain classifiers in a focused manner. The meta-classifiers use the data from the first group of domain classifiers to evaluate the query relative to the categories corresponding to the meta-classifiers. If the query corresponds to at least one of the meta-classifier categories, the query is assigned to the meta-classifier category with the highest ranking, probability, or other score. Preferably, the meta-classifiers are initially trained using a data set different from the set of training queries (or other training data) used for training a corresponding subject matter domain classifier and/or any of the first group of domain classifiers. Preferably, the meta-classifiers are based on a non-linear ensemble model for combining the information from the domain classifiers to generate a meta-classifier ranking, probability, or other score" teaches a plurality of meta-classifiers with a non-linear ensemble model corresponding to a domain (category) of a plurality of domains, with each meta-classifier having associated subject matter categories (classes) corresponding to its domain (category) and operating on corresponding domain classifier outputs (features) to evaluate the input with respect to the subject matter categories (classes) corresponding to its domain (category) (i.e. each meta-classifier has a corresponding domain (category) that is evaluated based on the corresponding categories (classes) having a same domain evaluation result (feature)). [0050]: "the initial features for the meta-classifiers are selected to be the output scores from the domain classifiers, plus various additional feature used by one or more of the domain classifiers" teaches that the domain output results from the domain classifiers (domain classifier outputs) are features for the meta-classifiers),
each accelerator is configured to perform: upon receiving of a data, making the computing circuit to execute the model for generating a classification result comprising at least one probability value to infer whether the data falls within the corresponding category (Fig. 4; Fig. 5; [0041]: "each meta-classifier can use the aggregated evaluation information to generate classification decision information for a query relative to the category for the meta-classifier. The meta-classifier can generate a ranking value, probability value, or other category score that indicates the association of a query with a subject matter area. The category scores from the meta-classifiers can then be compared. If none of the category scores is above a threshold value, then the query is not associated with any of the meta-classifier categories. If at least one of the category scores is above a threshold value, the query can be assigned to one or more meta-classifier categories, such as all categories with a score above the threshold value and/or all categories with a score within 90% of the highest category score. If a meta-classifier corresponding to a highest category score is associated with multiple domains, the outputs from the domain classifiers may be used to select one or more domains within the meta-classifier domains" teaches that each meta-classifier that receives the query (data) generates a probability value category (class) score (classification result comprising at least one probability value) that indicates which subject matter area (class) the query (data) falls within. The category (class) score is then used to determine if the query (data) falls within the corresponding domain (category)),
the plurality of accelerators are respectively corresponding to the plurality of categories ([0050]-[0051]: "the initial features for the meta-classifiers are selected to be the output scores from the domain classifiers, plus various additional feature used by one or more of the domain classifiers ... In order to train the meta-classifiers, the labeled queries in a training set are first evaluated by the domain classifiers (which have already been trained) to generate domain evaluation scores. The evaluation scores are then aggregated for use by each meta-classifier. In this example, the meta-classifiers are designed to provide a probability of association between a query and a meta-classifier category. Because some domains do not have a corresponding meta-classifier, queries with a marginal association should not necessarily be associated with a meta-classifier category. As a result, a meta-classifier threshold level is set at 50%. If a meta-classifier provides an association probability of lower than 50%, then the query is determined to not be associated with the corresponding meta-classifier category. If at least one category score is greater than 50%, the query is assigned to one or more meta-classifier categories. The meta-classifier with the highest probability (or other category score) is used to assign the query to the corresponding category. This continues for each meta-classifier with a category score greater than 50%. Optionally, if two or more of the meta-classifiers have a mutually exclusive relationship, a category assignment of the query to a first meta-classifier category is a second mutually exclusive category has already been assigned (such as based on a higher category score)" teaches that the meta-classifiers (accelerators) each corresponding to a category of the plurality of categories),
wherein the processor is configured to: establish the plurality of categories according to feature of each of a plurality of classes, wherein each of the plurality of categories includes at least one of the plurality of classes with same features (Fig. 4; Fig. 5; [0070]: "In FIG. 5, a query is evaluated 510 by a plurality of domain classifiers. The query is evaluated to determine a plurality of domain evaluation results. Category scores are generated 520 for meta-classifier categories based on the plurality of domain evaluation results. Preferably, the category scores are generated by meta-classifiers that are based on a non-linear classification model. In the embodiment shown in FIG. 5, at least one category score is greater than a threshold value. A meta-classifier category is then selected 530 based on the at least one category score that is greater than the threshold value. A domain is assigned 540 to the query based on the selected meta-classifier category" teaches that the meta-classifier generates category (class) scores based on domain evaluation results (features) and that the category (class) score that is greater than a threshold (at least one first class) is selected to assign (establish) a domain (category) having the category (class) score (i.e. a domain (category) is established based on selecting a category (class) score from the plurality of category (class) scores having a same domain evaluation result (feature)). [0050]-[0051]: "the initial features for the meta-classifiers are selected to be the output scores from the domain classifiers, plus various additional feature used by one or more of the domain classifiers ... In order to train the meta-classifiers, the labeled queries in a training set are first evaluated by the domain classifiers (which have already been trained) to generate domain evaluation scores. The evaluation scores are then aggregated for use by each meta-classifier. In this example, the meta-classifiers are designed to provide a probability of association between a query and a meta-classifier category. Because some domains do not have a corresponding meta-classifier, queries with a marginal association should not necessarily be associated with a meta-classifier category. As a result, a meta-classifier threshold level is set at 50%. If a meta-classifier provides an association probability of lower than 50%, then the query is determined to not be associated with the corresponding meta-classifier category. If at least one category score is greater than 50%, the query is assigned to one or more meta-classifier categories. The meta-classifier with the highest probability (or other category score) is used to assign the query to the corresponding category. This continues for each meta-classifier with a category score greater than 50%. Optionally, if two or more of the meta-classifiers have a mutually exclusive relationship, a category assignment of the query to a first meta-classifier category is a second mutually exclusive category has already been assigned (such as based on a higher category score)" teaches that the domain output results (domain evaluation results) from the domain classifiers are features for the meta-classifiers used to establish categories based on category (class) scores from the plurality of category (class) scores having a same domain evaluation result (feature))).
Marrero et al. and Li et al. are analogous to the claimed invention because they are directed to neural network classification.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate each accelerator being configured to store a model corresponding to one of a plurality of categories with at least one class having a same feature and being categorized within the category, each accelerator is configured to perform: upon receiving of a data, making the computing circuit to execute the model for generating a classification result comprising at least one probability value to infer whether the data falls within the corresponding category, the plurality of accelerators are respectively corresponding to the plurality of categories, wherein the processor is configured to: establish the plurality of categories according to feature of each of a plurality of classes, wherein each of the plurality of categories includes at least one of the plurality of classes with same features as taught by Li et al. to the disclosed invention of Marrero et al.
One of ordinary skill in the art would have been motivated to make this modification to provide "improved methods for assigning such queries to the correct category without requiring excessive additional resources" (Li et al. [0001]).
Marrero et al. in view of Li et al. does not appear to explicitly teach wherein the model is trained using a training dataset labeled with the at least one class.
However, Rossi et al. teaches wherein the model is trained using a training dataset labeled with the at least one class (Fig. 1; Fig. 2; [0115]-[0116]: "FIG. 1 presents a schematic overview of a classification task, in which an image is analyzed by a CNN to classify a photograph. After deployment, the classifier autonomously assigns the correct label to a test subject (in this case, to images). Alternatively, it could assign to each label a certain probability of membership to a class. The actual realization depends, of course, on the specific application and the classification method is chosen at design time. An example is depicted in FIG. 2, where the classifier processes the image of a cat. … There are various methods to realize a classifier with regard to how it performs classification. In a data-driven approach, the classifier undergoes a training process before its deployment. The training includes letting the classifier process a set of inputs whose ground truth is known a priori by the designer, the classification criteria being determined by a certain set of parameters" teaches that a set of inputs (training dataset) with known ground truths (labels for a plurality of classes) is used in the training process for a classifier model (model) (i.e. the classifier model is trained with labeled input data)).
Marrero et al. and Li et al. are analogous to the claimed invention because they are directed to neural network classification.
Rossi et al. is analogous to the claimed invention because it is directed to the implementation of a neural network accelerator.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate wherein the model is trained using a training dataset labeled with the at least one class as taught by Rossi et al. to the disclosed invention of Marrero et al. in view of Li et al.
One of ordinary skill in the art would have been motivated to make this modification "in order to determine the optimal hyper-parameters that provide the robustness of the classifier" (Rossi et al. [0132]) because "the results are compared with the ground truths to check for errors. A cost function is defined that evaluates the magnitude of these errors and provides a new set of parameters that improves the classification performance" (Rossi et al. [0117]).
Regarding Claim 7,
Marrero et al. in view of Li et al. and further in view of Rossi et al. teaches the electronic system of claim 6.
In addition, Marrero et al. further teaches wherein each of the plurality of accelerators comprises: the SRAM, configured to store the corresponding model (Fig. 7A; Fig. 8; [0080]: "In at least one embodiment, inference and/or training logic 715 may include, without limitation, a code and/or data storage 705 to store backward and/or output weight and/or input/output data corresponding to neurons or layers of a neural network trained and/or used for inferencing in aspects of one or more embodiments. In at least one embodiment, code and/or data storage 705 stores weight parameters and/or input/output data of each layer of a neural network trained or used in conjunction with one or more embodiments during backward propagation of input/output data and/or weight parameters during training and/or inferencing using aspects of one or more embodiments. … In at least one embodiment, code and/or data storage 705 may be cache memory, DRAM, SRAM, non-volatile memory (e.g., Flash memory), or other storage" teaches that each node computing resource 816 (accelerator) comprises inference and/or training logic 715, which comprises a data storage 705 that stores information for a training neural network (corresponding model), the data storage 705 being comprised of SRAM); and
the computing circuit, coupled to the SRAM, the computing circuit being configured to access the SRAM on the same chip internally in order to execute the corresponding model for generating the classification result upon receiving of the data (Fig. 7A; Fig. 8; [0082]: "In at least one embodiment, inference and/or training logic 715 may include, without limitation, one or more arithmetic logic unit(s) (“ALU(s)”) 710, including integer and/or floating point units, to perform logical and/or mathematical operations based, at least in part on, or indicated by, training and/or inference code (e.g., graph code), a result of which may produce activations (e.g., output values from layers or neurons within a neural network) stored in an activation storage 720 that are functions of input/output and/or weight parameter data stored in code and/or data storage 701 and/or code and/or data storage 705" teaches that each node computing resource 816 (accelerator) comprises inference and/or training logic 715, which comprises ALUs 710 (computing circuit) that performs operations of a neural network to generate outputs (executes corresponding model to generate classification result)).
Regarding Claim 8,
Marrero et al. in view of Li et al. and further in view of Rossi et al. teaches the electronic system of claim 6.
In addition, Rossi et al. further teaches wherein each accelerator is configured to generate the at least one probability value respectively corresponding to the at least one class within the corresponding category upon receiving of the data, for inferring which of the at least one class the received data falls within (Fig. 1; [0115]-[0116]: "FIG. 1 presents a schematic overview of a classification task, in which an image is analyzed by a CNN to classify a photograph. After deployment, the classifier autonomously assigns the correct label to a test subject (in this case, to images). Alternatively, it could assign to each label a certain probability of membership to a class. The actual realization depends, of course, on the specific application and the classification method is chosen at design time. An example is depicted in FIG. 2, where the classifier processes the image of a cat. … There are various methods to realize a classifier with regard to how it performs classification. In a data-driven approach, the classifier undergoes a training process before its deployment. The training includes letting the classifier process a set of inputs whose ground truth is known a priori by the designer, the classification criteria being determined by a certain set of parameters" teaches that an input image (training data) can be classified into different categories such as car, person, or animal. Fig. 2; [0115]-[0116]: "FIG. 1 presents a schematic overview of a classification task, in which an image is analyzed by a CNN to classify a photograph. After deployment, the classifier autonomously assigns the correct label to a test subject (in this case, to images). Alternatively, it could assign to each label a certain probability of membership to a class. The actual realization depends, of course, on the specific application and the classification method is chosen at design time. An example is depicted in FIG. 2, where the classifier processes the image of a cat. … There are various methods to realize a classifier with regard to how it performs classification. In a data-driven approach, the classifier undergoes a training process before its deployment. The training includes letting the classifier process a set of inputs whose ground truth is known a priori by the designer, the classification criteria being determined by a certain set of parameters" teaches that probabilities for classes within a category (at least one first probability value) are generated for the input image (training data), with the probabilities being inferred percentages of the object in the image being of the at least one class (e.g. for an image in the animal category, the probability of a cat class is 82% and of a dog class is 15% are classes within the category)).
Marrero et al. and Li et al. are analogous to the claimed invention because they are directed to neural network classification.
Rossi et al. is analogous to the claimed invention because it is directed to the implementation of a neural network accelerator.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate wherein each accelerator is configured to generate the at least one probability value respectively corresponding to the at least one class within the corresponding category upon receiving of the data, for inferring which of the at least one class the received data falls within as taught by Rossi et al. to the disclosed invention of Marrero et al. in view of Li et al.
One of ordinary skill in the art would have been motivated to make this modification "in order to determine the optimal hyper-parameters that provide the robustness of the classifier" (Rossi et al. [0132]) because "the results are compared with the ground truths to check for errors. A cost function is defined that evaluates the magnitude of these errors and provides a new set of parameters that improves the classification performance" (Rossi et al. [0117]).
Regarding Claim 9,
Marrero et al. in view of Li et al. and further in view of Rossi et al. teaches the electronic system of claim 8.
In addition, Rossi et al. further teaches wherein each classification result further comprises a category probability value, each accelerator is configured to generate the category probability value upon receiving of the data, for inferring whether the data falls within the category (Fig. 1; [0115]-[0116]: "FIG. 1 presents a schematic overview of a classification task, in which an image is analyzed by a CNN to classify a photograph. After deployment, the classifier autonomously assigns the correct label to a test subject (in this case, to images). Alternatively, it could assign to each label a certain probability of membership to a class. The actual realization depends, of course, on the specific application and the classification method is chosen at design time. An example is depicted in FIG. 2, where the classifier processes the image of a cat. … There are various methods to realize a classifier with regard to how it performs classification. In a data-driven approach, the classifier undergoes a training process before its deployment. The training includes letting the classifier process a set of inputs whose ground truth is known a priori by the designer, the classification criteria being determined by a certain set of parameters" teaches that an input image (training data) can be classified into different categories such as car, person, or animal. Fig. 2; [0115]-[0116]: "FIG. 1 presents a schematic overview of a classification task, in which an image is analyzed by a CNN to classify a photograph. After deployment, the classifier autonomously assigns the correct label to a test subject (in this case, to images). Alternatively, it could assign to each label a certain probability of membership to a class. The actual realization depends, of course, on the specific application and the classification method is chosen at design time. An example is depicted in FIG. 2, where the classifier processes the image of a cat. … There are various methods to realize a classifier with regard to how it performs classification. In a data-driven approach, the classifier undergoes a training process before its deployment. The training includes letting the classifier process a set of inputs whose ground truth is known a priori by the designer, the classification criteria being determined by a certain set of parameters" teaches that probabilities for classes outside of a category (category probability value) are generated for the input image (training data), with the probabilities being inferred percentages of the object in the image being outside of the category (e.g. for an image in the animal category, the probability of a hat class is 2% and of a mug class is 1% are classes outside the category, meaning that the percentage of the object not being in the category is 3%)).
Marrero et al. and Li et al. are analogous to the claimed invention because they are directed to neural network classification.
Rossi et al. is analogous to the claimed invention because it is directed to the implementation of a neural network accelerator.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate wherein each classification result further comprises a category probability value, each accelerator is configured to generate the category probability value upon receiving of the data, for inferring whether the data falls within the category as taught by Rossi et al. to the disclosed invention of Marrero et al. in view of Li et al.
One of ordinary skill in the art would have been motivated to make this modification "in order to determine the optimal hyper-parameters that provide the robustness of the classifier" (Rossi et al. [0132]) because "the results are compared with the ground truths to check for errors. A cost function is defined that evaluates the magnitude of these errors and provides a new set of parameters that improves the classification performance" (Rossi et al. [0117]).
Regarding Claim 10,
Marrero et al. in view of Li et al. and further in view of Rossi et al. teaches the electronic system of claim 9.
In addition, Rossi et al. further teaches wherein a summation of the category probability value and the at least one probability value of each classification result equals to 1 (Fig. 1; [0115]: "FIG. 1 presents a schematic overview of a classification task, in which an image is analyzed by a CNN to classify a photograph. After deployment, the classifier autonomously assigns the correct label to a test subject (in this case, to images). Alternatively, it could assign to each label a certain probability of membership to a class. The actual realization depends, of course, on the specific application and the classification method is chosen at design time. An example is depicted in FIG. 2, where the classifier processes the image of a cat" teaches that an image can be classified into different categories such as car, person, or animal. Fig. 2; [0115]: "FIG. 1 presents a schematic overview of a classification task, in which an image is analyzed by a CNN to classify a photograph. After deployment, the classifier autonomously assigns the correct label to a test subject (in this case, to images). Alternatively, it could assign to each label a certain probability of membership to a class. The actual realization depends, of course, on the specific application and the classification method is chosen at design time. An example is depicted in FIG. 2, where the classifier processes the image of a cat" teaches that the image can be assigned probabilities for classes within a category (at least one probability value) and for classes outside of the category (category probability value), with the summation of the probabilities of each classification result (cat is 0.82 and dog is 0.15 as classes within the animal category) with the category probability value (hat is 0.02 and mug is 0.01 for classes outside of the animal category) is equal to 1).
Marrero et al. and Li et al. are analogous to the claimed invention because they are directed to neural network classification.
Rossi et al. is analogous to the claimed invention because it is directed to the implementation of a neural network accelerator.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate wherein a summation of the category probability value and the at least one probability value of each classification result equals to 1 as taught by Rossi et al. to the disclosed invention of Marrero et al. in view of Li et al.
One of ordinary skill in the art would have been motivated to make this modification "in order to determine the optimal hyper-parameters that provide the robustness of the classifier" (Rossi et al. [0132]) because "the results are compared with the ground truths to check for errors. A cost function is defined that evaluates the magnitude of these errors and provides a new set of parameters that improves the classification performance" (Rossi et al. [0117]).
Regarding Claim 11,
Marrero et al. in view of Li et al. and further in view of Rossi et al. teaches the electronic system of claim 9.
In addition, Li et al. further teaches wherein the processor is configured to perform: upon receiving of the data, examining the category probability values generated by the plurality of accelerators to determine a selected category from the plurality of categories (Fig. 4; Fig. 5; [0016]: "A first group of query classifiers can be used to evaluate a query relative to various subject matter domains. This initial evaluation by the domain classifiers provides some type of ranking, probability, or other score for a query relative to the domains. The evaluation output or results from the first group of domain classifiers can then be used by a second group of meta-classifiers. The meta-classifiers are associated with subject matter categories that may correspond to a domain" teaches domain classifiers that upon receiving the query (data) generates domain (category) probabilities to determine which domain (category) the query (data) falls within); and
examining the at least one probability value corresponding to the selected category to determine which class the data falls within (Fig. 4; Fig. 5; [0041]: "After receiving the evaluation information from the domain classifiers, each meta-classifier can use the aggregated evaluation information to generate classification decision information for a query relative to the category for the meta-classifier. The meta-classifier can generate a ranking value, probability value, or other category score that indicates the association of a query with a subject matter area. The category scores from the meta-classifiers can then be compared. If none of the category scores is above a threshold value, then the query is not associated with any of the meta-classifier categories. If at least one of the category scores is above a threshold value, the query can be assigned to one or more meta-classifier categories, such as all categories with a score above the threshold value and/or all categories with a score within 90% of the highest category score. If a meta-classifier corresponding to a highest category score is associated with multiple domains, the outputs from the domain classifiers may be used to select one or more domains within the meta-classifier domains" teaches that the meta-classifier generates a ranking/probability category (class) score that indicates which subject matter area (class) within the domain (category) the query (data) falls within).
Marrero et al. and Li et al. are analogous to the claimed invention because they are directed to neural network classification.
Rossi et al. is analogous to the claimed invention because it is directed to the implementation of a neural network accelerator.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate wherein the processor is configured to perform: upon receiving of the data, examining the category probability values generated by the plurality of accelerators to determine a selected category from the plurality of categories; and examining the at least one probability value corresponding to the selected category to determine which class the data falls within as taught by Li et al. to the disclosed invention of Marrero et al. in view of Rossi et al.
One of ordinary skill in the art would have been motivated to make this modification to provide "improved methods for assigning such queries to the correct category without requiring excessive additional resources" (Li et al. [0001]).
Regarding Claim 12,
Marrero et al. in view of Li et al. and further in view of Rossi et al. teaches the electronic system of claim 9.
In addition, Li et al. further teaches wherein a category accelerator of the plurality of accelerators is configured to store a category model (Fig. 4; Fig. 5; [0016]: "A first group of query classifiers can be used to evaluate a query relative to various subject matter domains. This initial evaluation by the domain classifiers provides some type of ranking, probability, or other score for a query relative to the domains" teaches an initial domain classifier (category accelerator). [0027]: "Traditionally, domain classifiers are trained using a linear model, such as linear regression. In a linear model, a variety of features for the model are identified. A training set of data (such as queries) with known domain assignments is then used to fit weights for the features in the model. The weights are fit, for example, by using linear regression to determine weights that minimize the errors for the model over the training data set. This type of linear model approach appears to work in a satisfactory manner for a subject matter domain classifier" teaches that the domain classifier (category accelerator) comprises a linear model to classify domain assignments (linear model)), and
the category accelerator is configured to perform: upon receiving of the data, executing the category model for generating a plurality of category probability values respectively corresponding to the plurality of categories to infer which category the data falls within (Fig. 4; Fig. 5; [0016]: "A first group of query classifiers can be used to evaluate a query relative to various subject matter domains. This initial evaluation by the domain classifiers provides some type of ranking, probability, or other score for a query relative to the domains. The evaluation output or results from the first group of domain classifiers can then be used by a second group of meta-classifiers. The meta-classifiers are associated with subject matter categories that may correspond to a domain" teaches an initial domain classifier (category accelerator) that upon receiving the query (data) generates domain (category) probabilities to determine which domain (category) the query (data) falls within).
Marrero et al. and Li et al. are analogous to the claimed invention because they are directed to neural network classification.
Rossi et al. is analogous to the claimed invention because it is directed to the implementation of a neural network accelerator.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate wherein a category accelerator of the plurality of accelerators is configured to store a category model, and the category accelerator is configured to perform: upon receiving of the data, executing the category model for generating a plurality of category probability values respectively corresponding to the plurality of categories to infer which category the data falls within as taught by Li et al. to the disclosed invention of Marrero et al. in view of Rossi et al.
One of ordinary skill in the art would have been motivated to make this modification to provide "improved methods for assigning such queries to the correct category without requiring excessive additional resources" (Li et al. [0001]).
Regarding Claim 13,
Marrero et al. in view of Li et al. and further in view of Rossi et al. teaches the electronic system of claim 12.
In addition, Li et al. further teaches wherein after the category probability values are generated, the processor is configured to determine a selected category from the plurality of categories according to the category probability values (Fig. 4; Fig. 5; [0021]: "One option for combining domain classification scores is to assign a query to each domain that has a ranking or probability above a threshold value. Another option is to assign a query to the domain having the highest ranking, and to also assign the query to any additional domains with a ranking that is sufficiently close to the highest value, such as at least 90% of the highest value. Still another option is to assign a query to only one domain, such as by assigning a query to the domain with the highest ranking" teaches that the domain (category) with the highest ranking/probability from the generated domain classification scores is determined to be the assigned (selected) category for the query (data)).
Marrero et al. and Li et al. are analogous to the claimed invention because they are directed to neural network classification.
Rossi et al. is analogous to the claimed invention because it is directed to the implementation of a neural network accelerator.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate wherein after the category probability values are generated, the processor is configured to determine a selected category from the plurality of categories according to the category probability values as taught by Li et al. to the disclosed invention of Marrero et al. in view of Rossi et al.
One of ordinary skill in the art would have been motivated to make this modification to provide "improved methods for assigning such queries to the correct category without requiring excessive additional resources" (Li et al. [0001]).
Regarding Claim 14,
Marrero et al. in view of Li et al. and further in view of Rossi et al. teaches the electronic system of claim 13.
In addition, Li et al. further teaches wherein after the selected category is determined, the model corresponding to the selected category is configured to receive the data and generate at least one probability value respectively corresponding to at least one class within the selected category for inferring which class of the selected category the data falls within (Fig. 4; Fig. 5; [0016]: "The meta-classifiers are associated with subject matter categories that may correspond to a domain, or that may correspond to a plurality of domains. Because the meta-classifiers are limited in scope, the meta-classifiers can be trained to use the output from the domain classifiers in a focused manner. The meta-classifiers use the data from the first group of domain classifiers to evaluate the query relative to the categories corresponding to the meta-classifiers. If the query corresponds to at least one of the meta-classifier categories, the query is assigned to the meta-classifier category with the highest ranking, probability, or other score. Preferably, the meta-classifiers are initially trained using a data set different from the set of training queries (or other training data) used for training a corresponding subject matter domain classifier and/or any of the first group of domain classifiers. Preferably, the meta-classifiers are based on a non-linear ensemble model for combining the information from the domain classifiers to generate a meta-classifier ranking, probability, or other score. … In other embodiments, the meta-classifiers can be used to assign a query to a single category" teaches the meta classifier with the highest ranking/probability corresponding to the selected domain (category) is used to generate a meta-classifier category (class). Fig. 4; Fig. 5; [0041]: "each meta-classifier can use the aggregated evaluation information to generate classification decision information for a query relative to the category for the meta-classifier. The meta-classifier can generate a ranking value, probability value, or other category score that indicates the association of a query with a subject matter area. The category scores from the meta-classifiers can then be compared. If none of the category scores is above a threshold value, then the query is not associated with any of the meta-classifier categories. If at least one of the category scores is above a threshold value, the query can be assigned to one or more meta-classifier categories, such as all categories with a score above the threshold value and/or all categories with a score within 90% of the highest category score. If a meta-classifier corresponding to a highest category score is associated with multiple domains, the outputs from the domain classifiers may be used to select one or more domains within the meta-classifier domains" teaches that the meta-classifier generates a ranking/probability category (class) score that indicates which subject matter area (class) within the domain (category) the query (data) falls within).
Marrero et al. and Li et al. are analogous to the claimed invention because they are directed to neural network classification.
Rossi et al. is analogous to the claimed invention because it is directed to the implementation of a neural network accelerator.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate wherein after the selected category is determined, the model corresponding to the selected category is configured to receive the data and generate at least one probability value respectively corresponding to at least one class within the selected category for inferring which class of the selected category the data falls within as taught by Li et al. to the disclosed invention of Marrero et al. in view of Rossi et al.
One of ordinary skill in the art would have been motivated to make this modification to provide "improved methods for assigning such queries to the correct category without requiring excessive additional resources" (Li et al. [0001]).
Regarding Claim 15,
Marrero et al. teaches an operating method of an electronic system, comprising: providing a plurality of accelerators in the electronic system (Fig. 7A; Fig. 8; [0097]-[0098]: "Inference and/or training logic 715 are used to perform inferencing and/or training operations associated with one or more embodiments. Details regarding inference and/or training logic 715 are provided below in conjunction with FIGS. 7A and/or 7B. In at least one embodiment, inference and/or training logic 715 may be used in system FIG. 8 for inferencing or predicting operations based, at least in part, on weight parameters calculated using neural network training operations, neural network functions and/or architectures, or neural network use cases described herein. … one or more neural networks, as may include a feature extraction network and a classifier network" teaches a system for performing inferencing or predicting operations (operation method) using a neural network, including a classifier network. Fig. 8; [0088]: "In at least one embodiment, as shown in FIG. 8, data center infrastructure layer 810 may include a resource orchestrator 812, grouped computing resources 814, and node computing resources (“node C.R.s”) 816(1)-816(N), where “N” represents any whole, positive integer. In at least one embodiment, node C.R.s 816(1)-816(N) may include, but are not limited to, any number of central processing units (“CPUs”) or other processors (including accelerators, field programmable gate arrays (FPGAs), graphics processors, etc.), memory devices (e.g., dynamic read-only memory), storage devices (e.g., solid state or disk drives), network input/output (“NW I/O”) devices, network switches, virtual machines (“VMs”), power modules, and cooling modules, etc." teaches that the system comprises grouped computing resources 814 (processor), and node computing resources 816 (plurality of accelerators) for implementing the neural network),
respectively forming each of the plurality of accelerators by a computing circuit and a on-chip static random access memory (SRAM) on each of a plurality of chips (Fig. 7A; Fig. 8; [0080]-[0082]: "In at least one embodiment, inference and/or training logic 715 may include, without limitation, a code and/or data storage 705 to store backward and/or output weight and/or input/output data corresponding to neurons or layers of a neural network trained and/or used for inferencing in aspects of one or more embodiments. In at least one embodiment, code and/or data storage 705 stores weight parameters and/or input/output data of each layer of a neural network trained or used in conjunction with one or more embodiments during backward propagation of input/output data and/or weight parameters during training and/or inferencing using aspects of one or more embodiments. … In at least one embodiment, code and/or data storage 705 may be cache memory, DRAM, SRAM, non-volatile memory (e.g., Flash memory), or other storage … In at least one embodiment, inference and/or training logic 715 may include, without limitation, one or more arithmetic logic unit(s) (“ALU(s)”) 710, including integer and/or floating point units, to perform logical and/or mathematical operations based, at least in part on, or indicated by, training and/or inference code (e.g., graph code), a result of which may produce activations (e.g., output values from layers or neurons within a neural network) stored in an activation storage 720 that are functions of input/output and/or weight parameter data stored in code and/or data storage 701 and/or code and/or data storage 705" teaches that each node computing resource 816 (accelerator) comprises inference and/or training logic 715, which comprises a data storage 705 (memory), the data storage 705 being comprised of SRAM, and ALUs 710 (computing circuit) that performs operations of a neural network on a same chip).
Marrero et al. does not appear to explicitly teach each accelerator being configured to store a model corresponding to one of a plurality of categories with at least one class having a same feature and being categorized within the category, the model is trained using a training dataset labeled with the at least one class; and upon receiving of a data, making the computing circuit to execute, by each accelerator, the model for generating a classification result comprising at least one probability value to infer whether the data falls within the corresponding category; and establishing the plurality of categories, by the processor, according to feature of each of a plurality of classes, wherein each of the plurality of categories includes at least one of the plurality of classes with same features.
However, Li et al. teaches each accelerator being configured to store a model corresponding to one of a plurality of categories with at least one class having a same feature and being categorized within the category (Fig. 4; Fig. 5; [0016]: "A first group of query classifiers can be used to evaluate a query relative to various subject matter domains. This initial evaluation by the domain classifiers provides some type of ranking, probability, or other score for a query relative to the domains. The evaluation output or results from the first group of domain classifiers can then be used by a second group of meta-classifiers. The meta-classifiers are associated with subject matter categories that may correspond to a domain, or that may correspond to a plurality of domains. Because the meta-classifiers are limited in scope, the meta-classifiers can be trained to use the output from the domain classifiers in a focused manner. The meta-classifiers use the data from the first group of domain classifiers to evaluate the query relative to the categories corresponding to the meta-classifiers. If the query corresponds to at least one of the meta-classifier categories, the query is assigned to the meta-classifier category with the highest ranking, probability, or other score. Preferably, the meta-classifiers are initially trained using a data set different from the set of training queries (or other training data) used for training a corresponding subject matter domain classifier and/or any of the first group of domain classifiers. Preferably, the meta-classifiers are based on a non-linear ensemble model for combining the information from the domain classifiers to generate a meta-classifier ranking, probability, or other score" teaches a plurality of meta-classifiers with a non-linear ensemble model corresponding to a domain (category) of a plurality of domains, with each meta-classifier having associated subject matter categories (classes) corresponding to its domain (category) and operating on corresponding domain classifier outputs (features) to evaluate the input with respect to the subject matter categories (classes) corresponding to its domain (category) (i.e. each meta-classifier has a corresponding domain (category) that is evaluated based on the corresponding categories (classes) having a same domain evaluation result (feature)). [0050]: "the initial features for the meta-classifiers are selected to be the output scores from the domain classifiers, plus various additional feature used by one or more of the domain classifiers" teaches that the domain output results from the domain classifiers (domain classifier outputs) are features for the meta-classifiers); and
upon receiving of a data, making the computing circuit to execute, by each accelerator, the model for generating a classification result comprising at least one probability value to infer whether the data falls within the corresponding category (Fig. 4; Fig. 5; [0041]: "each meta-classifier can use the aggregated evaluation information to generate classification decision information for a query relative to the category for the meta-classifier. The meta-classifier can generate a ranking value, probability value, or other category score that indicates the association of a query with a subject matter area. The category scores from the meta-classifiers can then be compared. If none of the category scores is above a threshold value, then the query is not associated with any of the meta-classifier categories. If at least one of the category scores is above a threshold value, the query can be assigned to one or more meta-classifier categories, such as all categories with a score above the threshold value and/or all categories with a score within 90% of the highest category score. If a meta-classifier corresponding to a highest category score is associated with multiple domains, the outputs from the domain classifiers may be used to select one or more domains within the meta-classifier domains" teaches that each meta-classifier that receives the query (data) generates a probability value category (class) score (classification result comprising at least one probability value) that indicates which subject matter area (class) the query (data) falls within. The category (class) score is then used to determine if the query (data) falls within the corresponding domain (category)); and
establishing the plurality of categories, by the processor, according to feature of each of a plurality of classes, wherein each of the plurality of categories includes at least one of the plurality of classes with same features (Fig. 4; Fig. 5; [0070]: "In FIG. 5, a query is evaluated 510 by a plurality of domain classifiers. The query is evaluated to determine a plurality of domain evaluation results. Category scores are generated 520 for meta-classifier categories based on the plurality of domain evaluation results. Preferably, the category scores are generated by meta-classifiers that are based on a non-linear classification model. In the embodiment shown in FIG. 5, at least one category score is greater than a threshold value. A meta-classifier category is then selected 530 based on the at least one category score that is greater than the threshold value. A domain is assigned 540 to the query based on the selected meta-classifier category" teaches that the meta-classifier generates category (class) scores based on domain evaluation results (features) and that the category (class) score that is greater than a threshold (at least one first class) is selected to assign (establish) a domain (category) having the category (class) score (i.e. a domain (category) is established based on selecting a category (class) score from the plurality of category (class) scores having a same domain evaluation result (feature)). [0050]-[0051]: "the initial features for the meta-classifiers are selected to be the output scores from the domain classifiers, plus various additional feature used by one or more of the domain classifiers ... In order to train the meta-classifiers, the labeled queries in a training set are first evaluated by the domain classifiers (which have already been trained) to generate domain evaluation scores. The evaluation scores are then aggregated for use by each meta-classifier. In this example, the meta-classifiers are designed to provide a probability of association between a query and a meta-classifier category. Because some domains do not have a corresponding meta-classifier, queries with a marginal association should not necessarily be associated with a meta-classifier category. As a result, a meta-classifier threshold level is set at 50%. If a meta-classifier provides an association probability of lower than 50%, then the query is determined to not be associated with the corresponding meta-classifier category. If at least one category score is greater than 50%, the query is assigned to one or more meta-classifier categories. The meta-classifier with the highest probability (or other category score) is used to assign the query to the corresponding category. This continues for each meta-classifier with a category score greater than 50%. Optionally, if two or more of the meta-classifiers have a mutually exclusive relationship, a category assignment of the query to a first meta-classifier category is a second mutually exclusive category has already been assigned (such as based on a higher category score)" teaches that the domain output results (domain evaluation results) from the domain classifiers are features for the meta-classifiers used to establish categories based on category (class) scores from the plurality of category (class) scores having a same domain evaluation result (feature))).
Marrero et al. and Li et al. are analogous to the claimed invention because they are directed to neural network classification.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate each accelerator being configured to store a model corresponding to one of a plurality of categories with at least one class having a same feature and being categorized within the category; and upon receiving of a data, making the computing circuit to execute, by each accelerator, the model for generating a classification result comprising at least one probability value to infer whether the data falls within the corresponding category; and establishing the plurality of categories, by the processor, according to feature of each of a plurality of classes, wherein each of the plurality of categories includes at least one of the plurality of classes with same features as taught by Li et al. to the disclosed invention of Marrero et al.
One of ordinary skill in the art would have been motivated to make this modification to provide "improved methods for assigning such queries to the correct category without requiring excessive additional resources" (Li et al. [0001]).
Marrero et al. in view of Li et al. does not appear to explicitly teach the model is trained using a training dataset labeled with the at least one class.
However, Rossi et al. teaches the model is trained using a training dataset labeled with the at least one class (Fig. 1; Fig. 2; [0115]-[0116]: "FIG. 1 presents a schematic overview of a classification task, in which an image is analyzed by a CNN to classify a photograph. After deployment, the classifier autonomously assigns the correct label to a test subject (in this case, to images). Alternatively, it could assign to each label a certain probability of membership to a class. The actual realization depends, of course, on the specific application and the classification method is chosen at design time. An example is depicted in FIG. 2, where the classifier processes the image of a cat. … There are various methods to realize a classifier with regard to how it performs classification. In a data-driven approach, the classifier undergoes a training process before its deployment. The training includes letting the classifier process a set of inputs whose ground truth is known a priori by the designer, the classification criteria being determined by a certain set of parameters" teaches that a set of inputs (training dataset) with known ground truths (labels for a plurality of classes) is used in the training process for a classifier model (model) (i.e. the classifier model is trained with labeled input data)).
Marrero et al. and Li et al. are analogous to the claimed invention because they are directed to neural network classification.
Rossi et al. is analogous to the claimed invention because it is directed to the implementation of a neural network accelerator.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate the model is trained using a training dataset labeled with the at least one class as taught by Rossi et al. to the disclosed invention of Marrero et al. in view of Li et al.
One of ordinary skill in the art would have been motivated to make this modification "in order to determine the optimal hyper-parameters that provide the robustness of the classifier" (Rossi et al. [0132]) because "the results are compared with the ground truths to check for errors. A cost function is defined that evaluates the magnitude of these errors and provides a new set of parameters that improves the classification performance" (Rossi et al. [0117]).
Regarding Claim 16,
Marrero et al. in view of Li et al. and further in view of Rossi et al. teaches the operating method of claim 15.
In addition, Marrero et al. further teaches comprising: providing the SRAM configured to store the corresponding model (Fig. 7A; Fig. 8; [0080]: "In at least one embodiment, inference and/or training logic 715 may include, without limitation, a code and/or data storage 705 to store backward and/or output weight and/or input/output data corresponding to neurons or layers of a neural network trained and/or used for inferencing in aspects of one or more embodiments. In at least one embodiment, code and/or data storage 705 stores weight parameters and/or input/output data of each layer of a neural network trained or used in conjunction with one or more embodiments during backward propagation of input/output data and/or weight parameters during training and/or inferencing using aspects of one or more embodiments. … In at least one embodiment, code and/or data storage 705 may be cache memory, DRAM, SRAM, non-volatile memory (e.g., Flash memory), or other storage" teaches that each node computing resource 816 (accelerator) comprises inference and/or training logic 715, which comprises a data storage 705 that stores information for a training neural network (corresponding model), the data storage 705 being comprised of SRAM), and
the computing circuit in each accelerator, coupled to the SRAM and configured to access the SRAM to generate the classification result upon receiving of the data (Fig. 7A; Fig. 8; [0082]: "In at least one embodiment, inference and/or training logic 715 may include, without limitation, one or more arithmetic logic unit(s) (“ALU(s)”) 710, including integer and/or floating point units, to perform logical and/or mathematical operations based, at least in part on, or indicated by, training and/or inference code (e.g., graph code), a result of which may produce activations (e.g., output values from layers or neurons within a neural network) stored in an activation storage 720 that are functions of input/output and/or weight parameter data stored in code and/or data storage 701 and/or code and/or data storage 705" teaches that each node computing resource 816 (accelerator) comprises inference and/or training logic 715, which comprises ALUs 710 (computing circuit) that performs operations of a neural network to generate outputs (executes corresponding model to generate classification result)).
Regarding Claim 17,
Marrero et al. in view of Li et al. and further in view of Rossi et al. teaches the operating method of claim 15.
In addition, Rossi et al. further teaches wherein the operating method comprises: generating, by each accelerator, the at least one probability value respectively corresponding to the at least one class falls within the corresponding category upon receiving of the data, for inferring which of the at least one class the received data falls within (Fig. 1; [0115]-[0116]: "FIG. 1 presents a schematic overview of a classification task, in which an image is analyzed by a CNN to classify a photograph. After deployment, the classifier autonomously assigns the correct label to a test subject (in this case, to images). Alternatively, it could assign to each label a certain probability of membership to a class. The actual realization depends, of course, on the specific application and the classification method is chosen at design time. An example is depicted in FIG. 2, where the classifier processes the image of a cat. … There are various methods to realize a classifier with regard to how it performs classification. In a data-driven approach, the classifier undergoes a training process before its deployment. The training includes letting the classifier process a set of inputs whose ground truth is known a priori by the designer, the classification criteria being determined by a certain set of parameters" teaches that an input image (training data) can be classified into different categories such as car, person, or animal. Fig. 2; [0115]-[0116]: "FIG. 1 presents a schematic overview of a classification task, in which an image is analyzed by a CNN to classify a photograph. After deployment, the classifier autonomously assigns the correct label to a test subject (in this case, to images). Alternatively, it could assign to each label a certain probability of membership to a class. The actual realization depends, of course, on the specific application and the classification method is chosen at design time. An example is depicted in FIG. 2, where the classifier processes the image of a cat. … There are various methods to realize a classifier with regard to how it performs classification. In a data-driven approach, the classifier undergoes a training process before its deployment. The training includes letting the classifier process a set of inputs whose ground truth is known a priori by the designer, the classification criteria being determined by a certain set of parameters" teaches that probabilities for classes within a category (at least one first probability value) are generated for the input image (training data), with the probabilities being inferred percentages of the object in the image being of the at least one class (e.g. for an image in the animal category, the probability of a cat class is 82% and of a dog class is 15% are classes within the category)).
Marrero et al. and Li et al. are analogous to the claimed invention because they are directed to neural network classification.
Rossi et al. is analogous to the claimed invention because it is directed to the implementation of a neural network accelerator.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate wherein the operating method comprises: generating, by each accelerator, the at least one probability value respectively corresponding to the at least one class falls within the corresponding category upon receiving of the data, for inferring which of the at least one class the received data falls within as taught by Rossi et al. to the disclosed invention of Marrero et al. in view of Li et al.
One of ordinary skill in the art would have been motivated to make this modification "in order to determine the optimal hyper-parameters that provide the robustness of the classifier" (Rossi et al. [0132]) because "the results are compared with the ground truths to check for errors. A cost function is defined that evaluates the magnitude of these errors and provides a new set of parameters that improves the classification performance" (Rossi et al. [0117]).
Regarding Claim 18,
Marrero et al. in view of Li et al. and further in view of Rossi et al. teaches the operating method of claim 17.
In addition, Rossi et al. further teaches wherein each classification result further comprises a category probability value, the operating method comprises: generating, by each accelerator, the category probability value upon receiving of the data, for inferring whether the data falls within the category (Fig. 1; [0115]-[0116]: "FIG. 1 presents a schematic overview of a classification task, in which an image is analyzed by a CNN to classify a photograph. After deployment, the classifier autonomously assigns the correct label to a test subject (in this case, to images). Alternatively, it could assign to each label a certain probability of membership to a class. The actual realization depends, of course, on the specific application and the classification method is chosen at design time. An example is depicted in FIG. 2, where the classifier processes the image of a cat. … There are various methods to realize a classifier with regard to how it performs classification. In a data-driven approach, the classifier undergoes a training process before its deployment. The training includes letting the classifier process a set of inputs whose ground truth is known a priori by the designer, the classification criteria being determined by a certain set of parameters" teaches that an input image (training data) can be classified into different categories such as car, person, or animal. Fig. 2; [0115]-[0116]: "FIG. 1 presents a schematic overview of a classification task, in which an image is analyzed by a CNN to classify a photograph. After deployment, the classifier autonomously assigns the correct label to a test subject (in this case, to images). Alternatively, it could assign to each label a certain probability of membership to a class. The actual realization depends, of course, on the specific application and the classification method is chosen at design time. An example is depicted in FIG. 2, where the classifier processes the image of a cat. … There are various methods to realize a classifier with regard to how it performs classification. In a data-driven approach, the classifier undergoes a training process before its deployment. The training includes letting the classifier process a set of inputs whose ground truth is known a priori by the designer, the classification criteria being determined by a certain set of parameters" teaches that probabilities for classes outside of a category (category probability value) are generated for the input image (training data), with the probabilities being inferred percentages of the object in the image being outside of the category (e.g. for an image in the animal category, the probability of a hat class is 2% and of a mug class is 1% are classes outside the category, meaning that the percentage of the object not being in the category is 3%)).
Marrero et al. and Li et al. are analogous to the claimed invention because they are directed to neural network classification.
Rossi et al. is analogous to the claimed invention because it is directed to the implementation of a neural network accelerator.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate wherein each classification result further comprises a category probability value, the operating method comprises: generating, by each accelerator, the category probability value upon receiving of the data, for inferring whether the data falls within the category as taught by Rossi et al. to the disclosed invention of Marrero et al. in view of Li et al.
One of ordinary skill in the art would have been motivated to make this modification "in order to determine the optimal hyper-parameters that provide the robustness of the classifier" (Rossi et al. [0132]) because "the results are compared with the ground truths to check for errors. A cost function is defined that evaluates the magnitude of these errors and provides a new set of parameters that improves the classification performance" (Rossi et al. [0117]).
Regarding Claim 19,
Marrero et al. in view of Li et al. and further in view of Rossi et al. teaches the operating method of claim 18.
In addition, Rossi et al. further teaches wherein a summation of the category probability value and 25 the at least one probability value of each classification result equals to 1 (Fig. 1; [0115]: "FIG. 1 presents a schematic overview of a classification task, in which an image is analyzed by a CNN to classify a photograph. After deployment, the classifier autonomously assigns the correct label to a test subject (in this case, to images). Alternatively, it could assign to each label a certain probability of membership to a class. The actual realization depends, of course, on the specific application and the classification method is chosen at design time. An example is depicted in FIG. 2, where the classifier processes the image of a cat" teaches that an image can be classified into different categories such as car, person, or animal. Fig. 2; [0115]: "FIG. 1 presents a schematic overview of a classification task, in which an image is analyzed by a CNN to classify a photograph. After deployment, the classifier autonomously assigns the correct label to a test subject (in this case, to images). Alternatively, it could assign to each label a certain probability of membership to a class. The actual realization depends, of course, on the specific application and the classification method is chosen at design time. An example is depicted in FIG. 2, where the classifier processes the image of a cat" teaches that the image can be assigned probabilities for classes within a category (at least one probability value) and for classes outside of the category (category probability value), with the summation of the probabilities of each classification result (cat is 0.82 and dog is 0.15 as classes within the animal category) with the category probability value (hat is 0.02 and mug is 0.01 for classes outside of the animal category) is equal to 1).
Marrero et al. and Li et al. are analogous to the claimed invention because they are directed to neural network classification.
Rossi et al. is analogous to the claimed invention because it is directed to the implementation of a neural network accelerator.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate wherein a summation of the category probability value and 25 the at least one probability value of each classification result equals to 1 as taught by Rossi et al. to the disclosed invention of Marrero et al. in view of Li et al.
One of ordinary skill in the art would have been motivated to make this modification "in order to determine the optimal hyper-parameters that provide the robustness of the classifier" (Rossi et al. [0132]) because "the results are compared with the ground truths to check for errors. A cost function is defined that evaluates the magnitude of these errors and provides a new set of parameters that improves the classification performance" (Rossi et al. [0117]).
Regarding Claim 20,
Marrero et al. in view of Li et al. and further in view of Rossi et al. teaches the operating method of claim 18.
In addition, Li et al. further teaches comprising: upon receiving of the data, examining, by a processor, the category probability values generated by the plurality of accelerators to determine a selected category which the data falls within (Fig. 4; Fig. 5; [0016]: "A first group of query classifiers can be used to evaluate a query relative to various subject matter domains. This initial evaluation by the domain classifiers provides some type of ranking, probability, or other score for a query relative to the domains. The evaluation output or results from the first group of domain classifiers can then be used by a second group of meta-classifiers. The meta-classifiers are associated with subject matter categories that may correspond to a domain" teaches domain classifiers that upon receiving the query (data) generates domain (category) probabilities to determine which domain (category) the query (data) falls within); and
examining, by the processor, the at least one class probability value corresponding to the selected category to determine which class the data falls within (Fig. 4; Fig. 5; [0041]: "After receiving the evaluation information from the domain classifiers, each meta-classifier can use the aggregated evaluation information to generate classification decision information for a query relative to the category for the meta-classifier. The meta-classifier can generate a ranking value, probability value, or other category score that indicates the association of a query with a subject matter area. The category scores from the meta-classifiers can then be compared. If none of the category scores is above a threshold value, then the query is not associated with any of the meta-classifier categories. If at least one of the category scores is above a threshold value, the query can be assigned to one or more meta-classifier categories, such as all categories with a score above the threshold value and/or all categories with a score within 90% of the highest category score. If a meta-classifier corresponding to a highest category score is associated with multiple domains, the outputs from the domain classifiers may be used to select one or more domains within the meta-classifier domains" teaches that the meta-classifier generates a ranking/probability category (class) score that indicates which subject matter area (class) within the domain (category) the query (data) falls within).
Marrero et al. and Li et al. are analogous to the claimed invention because they are directed to neural network classification.
Rossi et al. is analogous to the claimed invention because it is directed to the implementation of a neural network accelerator.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate comprising: upon receiving of the data, examining, by a processor, the category probability values generated by the plurality of accelerators to determine a selected category which the data falls within; and examining, by the processor, the at least one class probability value corresponding to the selected category to determine which class the data falls within as taught by Li et al. to the disclosed invention of Marrero et al. in view of Rossi et al.
One of ordinary skill in the art would have been motivated to make this modification to provide "improved methods for assigning such queries to the correct category without requiring excessive additional resources" (Li et al. [0001]).
Response to Arguments
Applicant's arguments, filed 07/16/2026, with respect to the 35 U.S.C. 101 abstract idea rejections to claims 1-20 have been fully considered but they are not persuasive. Applicant asserts “Regarding independent claim 1, Applicant respectfully submits that claim 1 is at least eligible at Prong Two.
Step 2A Prong Two: The current amended claims 1, 6 and 15 recite a hardware structure of the accelerator, where the accelerator is formed by disposing a computing circuit with a on-chip SRAM in a same chip. That is, the current amended claims 1, 6 and 15 can be applied to a practical application and are not abstract ideas. Furthermore, even assuming, arguendo, that the claims are directed to an abstract idea, the claims integrate the alleged exception into a practical application and recite significantly more. As demonstrated in Example 3 of the USPTO's Subject Matter Eligibility Examples and supported by Enfish, claim elements that "improve the functioning of the claimed computer itself," such as allowing the computer to use less memory and achieve faster computation, integrate the abstract idea into a practical application. The amended claims specifically recite that each accelerator stores a category-specific model in its "on-chip SRAM" and the computing circuit "internally accesses the respective on-chip SRAM" to execute the model. As described in the specification [0030]-[0031], this specific physical anchoring and decentralized hardware configuration substantially reduces system complexity, decreases memory latency, and increases computation speed compared to conventional multi-core systems relying on DRAM. This aligns precisely with the USPTO's explicit recognition of machine learning technological improvements in Ex Parte Desjardins.
In summary, the rejection based on 35 U.S.C 101 to claims 1, 6, 15 should be withdrawn. The rejection based on 35 U.S.C 101 to claims 2-5, 7-14 and 17-20 respectively depend on claims 1, 6, 15 should be withdrawn, too” (Remarks Pages 8-9).
Examiner’s Response:
The examiner respectfully disagrees. Regarding claim 1, applicant has made general assertions that claim 1 recites claim elements that are not directed to an abstract idea and that even if the claim elements are directed to an abstract idea, the judicial exceptions are integrated into a practical application because the claims recite additional elements that cannot reasonably be characterized as covering mental processes or reflect an improvement to a technology or technical field. Regarding the “disposing a computing circuit and a memory on a same chip to form the accelerator” and “accessing the memory internally in the chip in order to execute the corresponding model by the computing circuit for generating the classification result upon receiving of the data” limitations of claim 1, these limitations, under their broadest reasonable interpretation, are high level recitation of applying a generic accelerator comprising generic SRAM and a generic computing circuit to execute a model to implement the abstract ideas such that it amounts to no more than merely using a computer as a tool to perform generic computer functions. See MPEP 2106.05(f): “Another consideration when determining whether a claim integrates a judicial exception into a practical application in Step 2A Prong Two or recites significantly more than a judicial exception in Step 2B is whether the additional elements amount to more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer. … Thus, for example, claims that amount to nothing more than an instruction to apply the abstract idea using a generic computer do not render an abstract idea eligible.”
Additionally, claim 1 further recites the abstract ideas of establishing a plurality of categories according to feature of each of a plurality of classes by the processor, wherein each of the plurality of categories includes at least one of the plurality of classes with same features (corresponds to evaluation and judgement; in particular, a human, with the assistance of pen and paper, can establish categories of classes with same features); and using at least one first class within one of the plurality of categories for verification (corresponds to evaluation and judgement; in particular, a human, with the assistance of pen and paper, can perform verification based on the first class within the first category). Since this limitation is directed to a judicial exception, it cannot provide any alleged solution or improvement. See MPEP 2106.05(a): “It is important to note, the judicial exception alone cannot provide the improvement. The improvement can be provided by one or more additional elements. See the discussion of Diamond v. Diehr, 450 U.S. 175, 187 and 191-92, 209 USPQ 1, 10 (1981)) in subsection II, below.”
Thus, it is the additional elements that are analyzed to determine whether the judicial exception is integrated into a practical application, not the judicial exception itself. The additional elements of claim 1 of “an accelerator”, “a processor”, “training a first model with the training dataset labeled with the plurality of classes”, “by the processor”, “disposing a computing circuit and a memory on a same chip to form the accelerator”, “implementing the first model on the accelerator by the computing circuit and the memory”, and “accessing the memory internally in the chip in order to execute the corresponding model by the computing circuit for generating the classification result upon receiving of the data”, as drafted, are additional elements that amount to no more than mere instructions to apply the exception for the abstract ideas. See MPEP 2106.05(f): “Another consideration when determining whether a claim integrates a judicial exception into a practical application in Step 2A Prong Two or recites significantly more than a judicial exception in Step 2B is whether the additional elements amount to more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer. … Thus, for example, claims that amount to nothing more than an instruction to apply the abstract idea using a generic computer do not render an abstract idea eligible.” Moreover, the recitation of “receiving a training dataset having a plurality of training data by a processor, wherein each training data is labeled to one of a plurality of classes”, as drafted, amounts to insignificant extra-solution activity. In particular, the additional elements corresponds to mere data gathering. See MPEP 2106.05(g). Accordingly, the additional elements do not integrate the abstract ideas into a practical application.
Furthermore, the claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, all of the additional elements are “mere instructions to apply an exception” (I.e. the additional elements describe a generic accelerator, processor, memory, computing circuit, model, and generic training of the model for applying the abstract ideas) or insignificant extra-solution activity (i.e. receiving data). Furthermore, the “receiving …” limitation is insignificant extra-solution activity that is well-understood, routine, and conventional according to MPEP 2106.05(d) (“The courts have recognized the following computer functions as well‐understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity… i. Receiving or transmitting data over a network). Mere instructions to apply an exception cannot provide an inventive concept. The claim is not patent eligible.
In other words, the limitations of “establishing a plurality of categories according to feature of each of a plurality of classes by the processor, wherein each of the plurality of categories includes at least one of the plurality of classes with same features” and “using at least one first class within one of the plurality of categories for verification” in claim 1 are abstract ideas that are directed to a judicial exception, so they cannot provide any alleged solution or improvement. Additionally, the limitations of “disposing a computing circuit and a memory on a same chip to form the accelerator” and “accessing the memory internally in the chip in order to execute the corresponding model by the computing circuit for generating the classification result upon receiving of the data” are additional elements that amount to no more than mere instructions to apply the exception for the abstract ideas, and the limitation of “receiving a training dataset having a plurality of training data by a processor, wherein each training data is labeled to one of a plurality of classes” is an additional element corresponding to insignificant extra-solution activity that is well-understood, routine, and conventional. Furthermore, the other additional elements recited in claim 1 are directed to mere instructions to apply an abstract idea. Therefore, claim 1 does not recite additional element(s) that can provide any alleged solution, improvement, or inventive concept.
Furthermore, regarding Appellant’s assertions regarding Ex Parte Desjardins, Examiner respectfully disagrees. The claims are distinguishable from Ex Parte Desjardins because Desjardins includes an improvement to machine learning related to addressing catastrophic forgetting while the current claims only use machine learning models at a high level to perform an abstract idea. In other words, Desjardins provided a specific training strategy that allows the model to preserve performance on earlier tasks even as it learns new ones, while the current claims merely employ a generic neural network and well-known backpropagation training of a generic accelerator and model at a high level for performing the abstract ideas. In addition, the current claims are similar to those discussed in Recentive Analytics, Inc. v. Fox Corp., et al. because the claims employ a high level model training process and the Recentive court has instructed that “using a machine learning technique … necessarily includes an iterative training step. Iterative training using selected training material and dynamic adjustments based on real-time changes are incident to the very nature of machine learning.” Furthermore, Recentive states “patents that do no more than claim the application of generic machine learning to new data environments, without disclosing improvements to the machine learning models to be applied, are patent ineligible under § 101.”
Applicant relies on the arguments above regarding independent claims 6 and 15 and dependent claims 2-5, 7-14, and 16-20 therefore the response above is applicable to those claims.
Applicant's arguments, filed 07/16/2026, with respect to the 35 U.S.C. 103 prior art rejections to claims 1-20 have been fully considered but they are not persuasive. Applicant asserts “Regarding independent claim 1, Applicant respectfully submits that "establishing a plurality of categories according to feature of each of a plurality of classes by the processor, wherein each of the plurality of categories includes at least one of the plurality of classes with same features" are at least not disclosed by the cited prior arts.
Please refer to Fig. 5 and [0070] of Li: …
Li discloses a query operation in Fig. 5. Li has disclosed to generate the category score in the query operation, and the meta-classifier category may be selected according to whether the category score is greater than the threshold value or not. That is, Li merely discloses to determine to select or deselect the meta-classifier category during the query operation.
On the contrary, currently presented claims 1, 6, 15 disclose "establishing a plurality of categories according to feature of each of a plurality of classes, wherein each of the plurality of categories includes at least one of the plurality of classes with same features". The classes to which training data are labeled are categorized by grouping the classes to a plurality of categories based on features of the classes, such that each category includes at least one of the classes.
On the other hand, currently presented claim I has disclosed that "training a first model with the training dataset labeled with the plurality of classes, and using at least one first class within one of the plurality of categories for verification by the processor". That is, currently presented claim 1 implements to simplify the first model by training the first model using at least one first class within one category. Correspondingly, computation speed of each of the accelerators can be enhanced because of the simplified model.
That is, Li fails to disclose currently presented claims 1, 6 and 1 5. On the other hand, Rossi and Schiff also fail to disclose currently presented claims 1, 6 and 15.
According to MPEP 2143.03, All Claim Limitations Must Be Considered [R- 08.2017]. "[a]ll words in a claim must be considered in judging the patentability of that claim against the prior art." In re Wilson, 424 F.2d 1382, 1385, 165 USPQ 494, 496 (CCPA 1970). "Examiners must consider all claim limitations when determining patentability of an invention over the prior art." In re Gulack, 703 F.2d 1381, 1385, 217 USPQ 401, 403-04 (Fed. Cir. 1983). Accordingly, withdrawal of the rejection under 35 U.S.C. 103 of independent claims 1, 6 and 15 is respectfully requested.
Regarding dependent claims 2-5, 7-14, and 16-20, these claims depend on the respective independent claims and overcome the obviousness rejections of the Office (MPEP 2143.03: "[i]f an independent claim is nonobvious under 35 U.S.C. 103, then any claim depending therefrom is nonobvious. In re Fine, 837 F.2d 1071, 5 USPQ2d 1596 (Fed. Cir. 1988)"). Accordingly, withdrawal of the rejections under 35 U.S.C. 103 of these claims is respectfully requested” (Remarks Pages 9-12).
Examiner’s Response:
The examiner respectfully disagrees. Regarding claim 1, the examiner respectfully disagrees to applicant’s assertion that “establishing a plurality of categories according to feature of each of a plurality of classes by the processor, wherein each of the plurality of categories includes at least one of the plurality of classes with same features” is at least not disclosed by the cited prior arts. In particular, examiner points to paragraphs [0070] and [0050]-[0051] of Li et al. (US 2014/0101119 A1), which specifically disclose, with respect to Fig. 4 and Fig. 5, "In FIG. 5, a query is evaluated 510 by a plurality of domain classifiers. The query is evaluated to determine a plurality of domain evaluation results. Category scores are generated 520 for meta-classifier categories based on the plurality of domain evaluation results. Preferably, the category scores are generated by meta-classifiers that are based on a non-linear classification model. In the embodiment shown in FIG. 5, at least one category score is greater than a threshold value. A meta-classifier category is then selected 530 based on the at least one category score that is greater than the threshold value. A domain is assigned 540 to the query based on the selected meta-classifier category" (i.e. the meta-classifier generates category (class) scores based on domain evaluation results (features) and that the category (class) score that is greater than a threshold (at least one first class) is selected to assign (establish) a domain (category) having the category (class) score (i.e. a domain (category) is established based on selecting a category (class) score from the plurality of category (class) scores having a same domain evaluation result (feature))) and "the initial features for the meta-classifiers are selected to be the output scores from the domain classifiers, plus various additional feature used by one or more of the domain classifiers ... In order to train the meta-classifiers, the labeled queries in a training set are first evaluated by the domain classifiers (which have already been trained) to generate domain evaluation scores. The evaluation scores are then aggregated for use by each meta-classifier. In this example, the meta-classifiers are designed to provide a probability of association between a query and a meta-classifier category. Because some domains do not have a corresponding meta-classifier, queries with a marginal association should not necessarily be associated with a meta-classifier category. As a result, a meta-classifier threshold level is set at 50%. If a meta-classifier provides an association probability of lower than 50%, then the query is determined to not be associated with the corresponding meta-classifier category. If at least one category score is greater than 50%, the query is assigned to one or more meta-classifier categories. The meta-classifier with the highest probability (or other category score) is used to assign the query to the corresponding category. This continues for each meta-classifier with a category score greater than 50%. Optionally, if two or more of the meta-classifiers have a mutually exclusive relationship, a category assignment of the query to a first meta-classifier category is a second mutually exclusive category has already been assigned (such as based on a higher category score)" (i.e. the domain output results (domain evaluation results) from the domain classifiers are features for the meta-classifiers used to establish categories based on category (class) scores from the plurality of category (class) scores having a same domain evaluation result (feature))). The category (class) scores from the plurality of category (class) scores having a same domain evaluation result (feature) that surpass the threshold value are selected to establish a domain (category) for the meta-classifier, with each meta-classifier corresponding to a different category.
Regarding the “training a first model with the training dataset labeled with the plurality of classes” limitation of claim 1, the examiner respectfully disagrees to applicant’s assertion that the limitation is at least not disclosed by the cited prior arts. In particular, examiner points to paragraphs [0115]-[0116] of Rossi et al. (US 2020/0310761 A1), which specifically disclose, with respect to Fig. 1 and Fig. 2, “FIG. 1 presents a schematic overview of a classification task, in which an image is analyzed by a CNN to classify a photograph. After deployment, the classifier autonomously assigns the correct label to a test subject (in this case, to images). Alternatively, it could assign to each label a certain probability of membership to a class. The actual realization depends, of course, on the specific application and the classification method is chosen at design time. An example is depicted in FIG. 2, where the classifier processes the image of a cat. … There are various methods to realize a classifier with regard to how it performs classification. In a data-driven approach, the classifier undergoes a training process before its deployment. The training includes letting the classifier process a set of inputs whose ground truth is known a priori by the designer, the classification criteria being determined by a certain set of parameters” (i.e. the set of inputs (training dataset) with known ground truths (labels for a plurality of classes) is used in the training process for a classifier model (first model) (i.e. the classifier model is trained with labeled input data)). The classifier model is trained using inputs with known ground truth labels for a plurality of classes.
Regarding the “using at least one first class within one of the plurality of categories for verification by the processor” limitation of claim 1, the examiner respectfully disagrees to applicant’s assertion that the limitation is at least not disclosed by the cited prior arts. In particular, examiner points to paragraphs [0075]-[0076] of Schiff et al. (US 2009/0172730 A1), which specifically disclose, with respect to Fig. 6B, "In one embodiment, the machine learning module 210 generates predetermined models for classifying images based on a hierarchical method. The hierarchical method fuses multiple binary classifiers into a k-classifier. In general, in a binary classifier, the more likely of two classes can be determined and thus selected. Thus, for multi-class classification, fusing of multiple binary classifiers can be performed. … In one embodiment, the k-classifier is built from results of a generalization accuracy computation. The generalization accuracy computation is, in one embodiment, determined from performing machine learning on a learning data set and recording the accuracy on a verification data set. For example, the training data set can be split into a learning data set and a verification data set. Machine learning can then be performed on the learning data set to generate a model. The accuracy of the model can then be determined by applying the model to the verification data set. By performing a learning process and a verification process, a probability for each pair of categories can be generated" (i.e. the model is trained and verified during the machine learning process by using the verification data set, which comprises accuracies/probabilities based on the classes (at least one first class) of the category (one of the plurality of categories))). A verification dataset comprising accuracies/probabilities based on the classes of the category is used to train and verify the machine learning classifier.
Applicant relies on the arguments above regarding independent claims 6 and 15 and dependent claims 2-5, 7-14, and 16-20 therefore the response above is applicable to those claims.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to BRIAN J HALES whose telephone number is (571)272-0878. The examiner can normally be reached M-F 9:00am - 5:00pm.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Kamran Afshar can be reached at (571) 272-7796. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/BRIAN J HALES/Examiner, Art Unit 2125
/KAMRAN AFSHAR/Supervisory Patent Examiner, Art Unit 2125
Prosecution Insights