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 .
Information Disclosure Statement
The information disclosure statements (IDS) submitted on September 21, 2023 and October 4, 2024 are in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statements are being considered by the examiner.
Specification
The abstract of the disclosure is objected to because of the following informalities:
“embodiments, A training” should read “embodiments, a training”
“initial weight and weight noise supplying unit” should read “initial weight, and a weight noise supplying unit”
A corrected abstract of the disclosure is required and must be presented on a separate sheet, apart from any other text. See MPEP § 608.01(b).
The disclosure is objected to because of the following informalities:
[0002]: “compensating a training operation variation” should read “compensating for training operation variation”
[0011]: “an PIM architecture which utilizes a next-generation non-volatile memory devices” should read “a PIM architecture which utilizes a next-generation non-volatile memory device”; “are in the spotlight” should read “is in the spotlight”
[0012]: “a process variations” should read “a process variation”
[0023]: “a weight noise supplying” should read “a weight noise supplying unit”; “based on the comparison result performed in the accuracy comparing unit” should read “based on a result of the comparison performed in the accuracy comparing unit”
[0024]: “by the weight copying unit” should read “by the weight copy engine”
[0028]: “an calibration weight” should read “a calibration weight”
[0031]: “weight noise supplying to apply” should read “weight noise supplying unit to apply”; “based on a comparison result performed in the accuracy comparing unit” should read “based on a result of the comparison performed in the accuracy comparing unit”
[0032]: “the weight copying unit” should read “the weight copy engine”
[0033]: “by applying a nose” should read “by applying a noise”
[0034]: “When” should read “when”
[0035]: “an calibration weight” should read “a calibration weight”
[0037]: “any one of the training operation variation compensating method” should read “any one of the training operation variation compensating methods”
[0042]: “training operation variation of an artificial neural network” should read “training operation variation compensating apparatus of an artificial neural network”
[0045]: “ordinary skilled in the art” should read “ordinary skill in the art”
[0051]: “system, an” should read “system, and an”
[0056]: “applies to any” should read “applies noise to any”
[0061]: “applies a nose” should read “applies a noise”
[0063]: “wo or three” should read “two or three”
[0067]: “applies to the noise” should read “applies the noise”
[0080]: “an calibration weight” should read “a calibration weight”
[0086]: “and in another exemplary embodiments and some blocks” should read “and in other exemplary embodiments some blocks”; “variant compensating” should read “variation compensating”
[0091]: “an connection intensity” should read “a connection intensity”; “each neural” should read “each neuron”
[0097]: “training operation variation of an artificial neural network” should read “training operation variation compensating apparatus of an artificial neural network”
[0122]: “applies to any” should read “applies noise to any”
[0124]: “a comparison result performed in the accuracy comparing unit” should read “a result of the comparison performed in the accuracy comparing unit”
[0134]: “compensating a training operation” should read “compensating for training operation”
Appropriate correction is required.
Claim Objections
Claims 1-15 are objected to because of the following informalities:
Claim 1: "a weight noise supplying" should read "a weight noise supplying unit"; “training an artificial neural network” should read “training the artificial neural network”
Claim 5: "performs" should read "perform"
Claim 6: "an calibration weight" should read "a calibration weight"
Claim 9: "allowing weight noise supplying" should read "allowing the weight noise supplying unit"; "a comparison result performed" should read "a result of the comparison performed"
Claim 10: “allowing a weight copy engine” should read “allowing the weight copy engine”
Claim 11: "weight noise supplying" should read "the weight noise supplying unit"
Claim 12: "the comparison result performed" should read "the result of the comparison performed"
Claim 13: "an calibration weight" should read "a calibration weight"
Claims 2-4, 7-8, and 14-15 are objected to due to dependency on an objected-to base claim.
Appropriate correction is required.
Claim Interpretation
The following is a quotation of 35 U.S.C. 112(f):
(f) Element in Claim for a Combination. – An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof.
The following is a quotation of pre-AIA 35 U.S.C. 112, sixth paragraph:
An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof.
The claims in this application are given their broadest reasonable interpretation using the plain meaning of the claim language in light of the specification as it would be understood by one of ordinary skill in the art. The broadest reasonable interpretation of a claim element (also commonly referred to as a claim limitation) is limited by the description in the specification when 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is invoked.
As explained in MPEP § 2181, subsection I, claim limitations that meet the following three-prong test will be interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph:
(A) the claim limitation uses the term “means” or “step” or a term used as a substitute for “means” that is a generic placeholder (also called a nonce term or a non-structural term having no specific structural meaning) for performing the claimed function;
(B) the term “means” or “step” or the generic placeholder is modified by functional language, typically, but not always linked by the transition word “for” (e.g., “means for”) or another linking word or phrase, such as “configured to” or “so that”; and
(C) the term “means” or “step” or the generic placeholder is not modified by sufficient structure, material, or acts for performing the claimed function.
Use of the word “means” (or “step”) in a claim with functional language creates a rebuttable presumption that the claim limitation is to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites sufficient structure, material, or acts to entirely perform the recited function.
Absence of the word “means” (or “step”) in a claim creates a rebuttable presumption that the claim limitation is not to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is not interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites function without reciting sufficient structure, material or acts to entirely perform the recited function.
Claim limitations in this application that use the word “means” (or “step”) are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action. Conversely, claim limitations in this application that do not use the word “means” (or “step”) are not being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action.
This application includes one or more claim limitations that do not use the word “means,” but are nonetheless being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, because the claim limitation(s) uses a generic placeholder that is coupled with functional language without reciting sufficient structure to perform the recited function and the generic placeholder is not preceded by a structural modifier. Such claim limitation(s) is/are: “weight noise supplying unit”; “accuracy comparing unit”; and “weight copying unit”.
Because this/these claim limitation(s) is/are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, it/they is/are being interpreted to cover the corresponding structure described in the specification as performing the claimed function, and equivalents thereof.
If applicant does not intend to have this/these limitation(s) interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, applicant may: (1) amend the claim limitation(s) to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph (e.g., by reciting sufficient structure to perform the claimed function); or (2) present a sufficient showing that the claim limitation(s) recite(s) sufficient structure to perform the claimed function so as to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph.
Claim Rejections - 35 USC § 112
The following is a quotation of the first paragraph of 35 U.S.C. 112(a):
(a) IN GENERAL.—The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor or joint inventor of carrying out the invention.
The following is a quotation of the first paragraph of pre-AIA 35 U.S.C. 112:
The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor of carrying out his invention.
Claims 1-15 are rejected under 35 U.S.C. 112(a) or 35 U.S.C. 112 (pre-AIA ), first paragraph, as failing to comply with the written description requirement. The claim(s) contains subject matter which was not described in the specification in such a way as to reasonably convey to one skilled in the relevant art that the inventor or a joint inventor, or for applications subject to pre-AIA 35 U.S.C. 112, the inventor(s), at the time the application was filed, had possession of the claimed invention.
Claim 1 recites the limitation “accuracy comparing unit” which invokes 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. However, the written description fails to disclose the corresponding structure, material, or acts for performing the entire claimed functions and to clearly link the structure, material, or acts to the functions. Therefore, the written description is inadequate to show that the inventor had possession of the claimed invention at the time of filing. Claims 2-8 are rejected for being dependent on a rejected base claim. See rejections under 35 U.S.C. 112(b) below for further analysis.
Claims 2 and 10 recite the limitation “weight copying unit” which invokes 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. However, the written description fails to disclose the corresponding structure, material, or acts for performing the entire claimed functions and to clearly link the structure, material, or acts to the functions. Therefore, the written description is inadequate to show that the inventor had possession of the claimed invention at the time of filing. Claims 3-5 and 11-12 are rejected for being dependent on a rejected base claim. See rejections under 35 U.S.C. 112(b) below for further analysis.
Claim 9 recites the limitations “weight noise supplying unit” and “accuracy comparing unit” which invoke 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. However, the written description fails to disclose the corresponding structure, material, or acts for performing the entire claimed functions and to clearly link the structure, material, or acts to the functions. Therefore, the written description is inadequate to show that the inventor had possession of the claimed invention at the time of filing. Claims 10-15 are rejected for being dependent on a rejected base claim. See rejections under 35 U.S.C. 112(b) below for further analysis.
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 1-15 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 “accuracy comparing unit” which invokes 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. However, the written description fails to disclose the corresponding structure, material, or acts for performing the entire claimed functions and to clearly link the structure, material, or acts to the functions. Therefore, the claim is indefinite and is rejected under 35 U.S.C. 112(b) or pre-AIA 35 U.S.C. 112, second paragraph. For examination purposes, Examiner is interpreting “accuracy comparing unit” to be any software that performs the function of accuracy comparing. Claims 2-8 are additionally rejected for being dependent on a rejected base claim.
Applicant may:
(a) Amend the claim so that the claim limitation will no longer be interpreted as a limitation under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph;
(b) Amend the written description of the specification such that it expressly recites what structure, material, or acts perform the entire claimed function, without introducing any new matter (35 U.S.C. 132(a)); or
(c) Amend the written description of the specification such that it clearly links the structure, material, or acts disclosed therein to the function recited in the claim, without introducing any new matter (35 U.S.C. 132(a)).
If applicant is of the opinion that the written description of the specification already implicitly or inherently discloses the corresponding structure, material, or acts and clearly links them to the function so that one of ordinary skill in the art would recognize what structure, material, or acts perform the claimed function, applicant should clarify the record by either:
(a) Amending the written description of the specification such that it expressly recites the corresponding structure, material, or acts for performing the claimed function and clearly links or associates the structure, material, or acts to the claimed function, without introducing any new matter (35 U.S.C. 132(a)); or
(b) Stating on the record what the corresponding structure, material, or acts, which are implicitly or inherently set forth in the written description of the specification, perform the claimed function. For more information, see 37 CFR 1.75(d) and MPEP §§ 608.01(o) and 2181.
Claim 1 recites “the comparison result performed in the accuracy comparing unit” which has insufficient antecedent basis in the claims. Examiner recommends amending this limitation to read “a result of the comparison performed in the accuracy comparing unit” to overcome this rejection. Claims 2-8 are rejected due to dependency on claim 1.
Claim 1 recites “a robust memory array.” The term “robust” is a relative term which renders the claim indefinite. The term “robust” is not defined by the claim, the specification does not provide a standard for ascertaining the requisite degree, and one of ordinary skill in the art would not be reasonably apprised of the scope of the invention. For examination purposes, Examiner is interpreting “robust memory array” to be any memory array. Claims 2-8 are rejected due to dependency on claim 1.
Claim 2 and 10 recite the limitation “weight copying unit” which invokes 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. However, the written description fails to disclose the corresponding structure, material, or acts for performing the entire claimed functions and to clearly link the structure, material, or acts to the functions. Therefore, the claim is indefinite and is rejected under 35 U.S.C. 112(b) or pre-AIA 35 U.S.C. 112, second paragraph. Claims 3-5 and 11-12 are rejected for being dependent on a rejected base claim.
Claims 2 and 10 recite “the weight copying unit” which has insufficient antecedent basis in the claims. Examiner recommends amending this limitation to read “the weight copy engine” to overcome this rejection. For examination purposes, Examiner is interpreting “the weight copying unit” to be “the weight copy engine.” Claims 3-5 are rejected due to dependency on claim 2, and claims 11-12 are rejected due to dependency on claim 10.
Claim 3 recites “the weight noise supplying unit” which has insufficient antecedent basis in the claims. Examiner recommends amending “a weight noise supplying” in claim 1 to read “a weight noise supplying unit” in order to establish proper antecedent basis and overcome this rejection. Claims 4-5 are rejected due to dependency on claim 3.
Claim 9 recites the limitations “weight noise supplying unit” and “accuracy comparing unit” which invoke 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. However, the written description fails to disclose the corresponding structure, material, or acts for performing the entire claimed function and to clearly link the structure, material, or acts to the function. Therefore, the claim is indefinite and is rejected under 35 U.S.C. 112(b) or pre-AIA 35 U.S.C. 112, second paragraph. For examination purposes, Examiner is interpreting “weight noise supplying unit” is to be any software that performs the function of weight noise supplying, and “accuracy comparing unit” to be any software that performs the function of accuracy comparing. Claims 10-15 are rejected for being dependent on a rejected base claim.
Claim 9 recites “a robust memory array.” The term “robust” is a relative term which renders the claim indefinite. The term “robust” is not defined by the claim, the specification does not provide a standard for ascertaining the requisite degree, and one of ordinary skill in the art would not be reasonably apprised of the scope of the invention. For examination purposes, Examiner is interpreting “robust memory array” to be any memory array. Claims 10-15 are rejected due to dependency on claim 9.
Claim 14 recites “when the robust memory array performs an operation for training the artificial neural network like the non-volatile memory array.” The term “like” is a relative term which renders the claim indefinite. The term “like” is not defined by the claim, the specification does not provide a standard for ascertaining the requisite degree, and one of ordinary skill in the art would not be reasonably apprised of the scope of the invention. Examiner recommends amending this limitation to read “when the robust memory array performs an operation for training the artificial neural network which is the same as the operation performed by the non-volatile memory array” to overcome this rejection.
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.
Claim 15 is rejected under 35 U.S.C. 101 because the claimed invention is directed to non-statutory subject matter. The claim does not fall within at least one of the four categories of patent eligible subject matter because it is directed to a computer program, and is therefore directed to software per se. Examiner recommends amending the claim to read “A non-transitory computer readable recording medium storing a computer program to allow a computer to execute the training operation variation compensating method of an artificial neural network according to claim 9” in order to overcome this rejection.
Claims 1-15 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. The analysis of the claims will follow the 2019 Revised Patent Subject Matter Eligibility Guidance (“2019 PEG”).
Claim 1
Step 1: The claim recites an apparatus, and therefore is directed to the statutory category of machines.
Step 2A Prong 1: The claim recites, inter alia:
“a non-volatile memory array which performs an operation… using a predetermined initial weight”; This limitation encompasses, excepting the recitation of generic computer components (a non-volatile memory array), mentally performing an operation using a predetermined initial weight, such as mentally multiplying an input value by the predetermined initial weight.
“a robust memory array which performs an operation… which is the same as the non-volatile memory array using the initial weight”; This limitation encompasses, excepting the recitation of generic computer components (a robust memory array), mentally performing an operation which is the same as the non-volatile memory array using the initial weight, such as multiplying an input value by the initial weight.
“a weight noise supplying which applies a noise… to perform the… operation using a weight obtained by reflecting the noise to the initial weight”; This limitation encompasses, excepting the recitation of generic computer components (a weight noise supplying [unit]), mentally applying a noise by reflecting the noise to the initial weight (such as by adding the noise value to the initial weight value) to perform the operation using the new weight.
“an accuracy comparing unit which compares a first accuracy for a value calculated in the non-volatile memory array and a second accuracy for a value calculated in the robust memory array”; This limitation encompasses, excepting the recitation of generic computer components (an accuracy comparing unit), mentally comparing a first accuracy for a value calculated in the non-volatile memory array and a second accuracy for a value calculated in the robust memory array.
“a weight calibrator which performs backward propagation based on the comparison result performed in the accuracy comparing unit”; This limitation encompasses, excepting the recitation of generic computer components (a weight calibrator), a mathematical calculation of performing backward propagation based on the comparison result performed in the accuracy comparing unit.
Step 2A Prong 2: The judicial exception is not integrated into a practical application. The claim further recites “a non-volatile memory array,” “a robust memory array,” “a weight noise supplying [unit],” “an accuracy comparing unit,” and “a weight calibrator,” for performing the method steps, and that the noise is applied “to any one of the non-volatile memory array or the robust memory array,” however these limitations amount to mere instructions to apply an exception using a generic computer programmed with a generic class of computer algorithms (MPEP 2106.05(f)). The claim additionally recites that the operations are for training an artificial neural network, however these limitations amount to merely generally linking the use of a judicial exception to the field of use of neural network training (MPEP 2106.05(h)).
Step 2B: The claim does not contain significantly more than the judicial exception. The analysis at this step mirrors that of Step 2A Prong 2 above. As an ordered whole, the claim is directed to a mentally performable process of performing an operation using a predetermined initial weight, applying noise to the initial weight to perform the operation using a weight obtained by reflecting the noise to the initial weight, comparing first and second accuracies, and performing backward propagation based on the result of the comparison. Nothing in the claim provides significantly more than this. As such, the claim is not patent eligible.
Claim 2
Step 1: A machine, as above.
Step 2A Prong 1: The claim recites, inter alia:
“a weight copy engine which performs weight copy… based on the comparison result…”; This limitation encompasses, excepting the recitation of generic computer components (a weight copy engine), mentally performing weight copy based on the comparison result.
Step 2A Prong 2: The judicial exception is not integrated into a practical application. The claim further recites that the weight copy is performed by “a weight copy engine” and “from a memory array having a higher accuracy to a memory array having a lower accuracy,” however these limitations amount to mere instructions to apply an exception using a generic computer programmed with a generic class of computer algorithms (MPEP 2106.05(f)). The claim additionally recites that the comparison result is “obtained from the accuracy comparing unit, by the weight copying unit,” however this limitation amounts to the insignificant extra-solution activity of mere data gathering and outputting (MPEP 2106.05(g)).
Step 2B: The claim does not contain significantly more than the judicial exception. The “comparison result obtained from the accuracy comparing unit, by the weight copying unit” limitation, in addition to being insignificant extra solution activity, is also directed to the well-understood, routine, and conventional activity of receiving or transmitting data over a network (MPEP 2106.05(d)(II)(i) buySAFE, Inc. v. Google, Inc., 765 F.3d 1350, 1355, 112 USPQ2d 1093, 1096 (Fed. Cir. 2014) (computer receives and sends information over a network)). The further analysis at this step mirrors that of Step 2A Prong 2 above.
Claim 3
Step 1: A machine, as above.
Step 2A Prong 1: The claim recites, inter alia:
“wherein the weight noise supplying unit applies a noise…at every predetermined first epoch to perform the… operation using the weight in which the noise is reflected”; This limitation encompasses, excepting the recitation of generic computer components (the weight noise supplying unit), mentally applying a noise at every predetermined first epoch by reflecting the noise to a weight (such as by adding the noise value to the weight value) to perform the operation using the new weight.
Step 2A Prong 2: The judicial exception is not integrated into a practical application. The claim further recites that the “weight noise supplying unit” performs the limitation, and that the noise is applied “to the robust memory array,” however these limitations amount to mere instructions to apply an exception using a generic computer programmed with a generic class of computer algorithms (MPEP 2106.05(f)). The claim additionally recites that the operation is a “training” operation, however this limitation amounts to merely generally linking the use of a judicial exception to the field of use of neural network training (MPEP 2106.05(h)).
Step 2B: The claim does not contain significantly more than the judicial exception. The analysis at this step mirrors that of Step 2A Prong 2 above.
Claim 4
Step 1: A machine, as above.
Step 2A Prong 1: The claim recites, inter alia:
“wherein when the first accuracy is lower than the second accuracy, the weight copy engine performs the weight copy…”; This limitation encompasses, excepting the recitation of generic computer components (the weight copy engine), mentally performing the weight copy.
“the weight calibrator performs the backward propagation based on a training operation performed in the robust memory array”; This limitation encompasses, excepting the recitation of generic computer components (the weight calibrator), a mathematical calculation of performing the backward propagation based on a training operation performed in the robust memory array.
Step 2A Prong 2: The judicial exception is not integrated into a practical application. The claim further recites that the “weight copy engine” and “the weight calibrator” perform the limitations, and that the weight copy is performed “from the robust memory array to the non-volatile memory array” however these limitations amount to mere instructions to apply an exception using a generic computer programmed with a generic class of computer algorithms (MPEP 2106.05(f)).
Step 2B: The claim does not contain significantly more than the judicial exception. The analysis at this step mirrors that of Step 2A Prong 2 above.
Claim 5
Step 1: A machine, as above.
Step 2A Prong 1: The claim recites, inter alia:
“wherein when the first accuracy is equal to or higher than the second accuracy, the weight copy engine does not copy the weight and the weight calibrator does not performs the backward propagation”; This limitation encompasses, excepting the recitation of generic computer components (the weight copy engine and the weight calibrator), mentally deciding not to copy the weight and not to perform the backward propagation when the first accuracy is equal to or higher than the second accuracy.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. The claim further recites “the weight copy engine” and “the weight calibrator,” however these limitations amount to mere instructions to apply an exception using a generic computer programmed with a generic class of computer algorithms (MPEP 2106.05(f)).
Step 2B: The claim does not contain significantly more than the judicial exception. The analysis at this step mirrors that of Step 2A Prong 2 above.
Claim 6
Step 1: A machine, as above.
Step 2A Prong 1: The claim recites, inter alia:
“wherein the accuracy comparing unit compares the first accuracy and the second accuracy at every predetermined second reference epoch, and when the first accuracy is lower than the second accuracy a predetermined reference epoch or more, the weight calibrator applies an calibration weight… to perform the… operation using the calibration weight”; This limitation encompasses, excepting the recitation of generic computer components (the accuracy comparing unit and the weight calibrator), mentally comparing the first accuracy and the second accuracy at every predetermined second reference epoch, and when the first accuracy is lower than the second accuracy a predetermined reference epoch or more, applying a calibration weight to perform the operation using the calibration weight.
Step 2A Prong 2: The judicial exception is not integrated into a practical application. The claim further recites that the steps are performed by “the accuracy comparing unit” and “the weight calibrator,” and the calibration weight is applied to “allow the non-volatile memory array to perform the training operation,” however these limitations amount to mere instructions to apply an exception using a generic computer programmed with a generic class of computer algorithms (MPEP 2106.05(f)). The claim additionally recites that the operation is a “training” operation, however this limitation amounts to merely generally linking the use of a judicial exception to the field of use of neural network training (MPEP 2106.05(h)).
Step 2B: The claim does not contain significantly more than the judicial exception. The analysis at this step mirrors that of Step 2A Prong 2 above.
Claim 7
Step 1: A machine, as above.
Step 2A Prong 1: The claim recites the same judicial exception as claim 1.
Step 2A Prong 2: The judicial exception is not integrated into a practical application. The claim further recites “wherein the non-volatile memory array is based on at least one of a magnetoresistive random access memory (MRAM), a phase change memory, and a ferroelectric random access memory (FeRAM),” however this limitation amounts to mere instructions to apply an exception using a generic computer programmed with a generic class of computer algorithms (MPEP 2106.05(f)).
Step 2B: The claim does not contain significantly more than the judicial exception. The analysis at this step mirrors that of Step 2A Prong 2 above.
Claim 8
Step 1: A machine, as above.
Step 2A Prong 1: The claim recites the same judicial exception as claim 1.
Step 2A Prong 2: The judicial exception is not integrated into a practical application. The claim further recites “wherein the robust memory array is based on a static random access memory (SRAM),” however this limitation amounts to mere instructions to apply an exception using a generic computer programmed with a generic class of computer algorithms (MPEP 2106.05(f)).
Step 2B: The claim does not contain significantly more than the judicial exception. The analysis at this step mirrors that of Step 2A Prong 2 above.
Claim 9
Step 1: The claim recites a method, and therefore is directed to the statutory category of processes.
Step 2A Prong 1: The claim recites, inter alia:
“a step of allowing the non-volatile memory array and the robust memory array to perform an operation… using a predetermined initial weight”; This limitation encompasses, excepting the recitation of generic computer components (the non-volatile memory array and the robust memory array), mentally performing an operation using a predetermined initial weight, such as mentally multiplying an input value by the predetermined initial weight.
“a step of allowing weight noise supplying to apply a noise… to perform the… operation using a weight obtained by reflecting the noise to the initial weight”; This limitation encompasses, excepting the recitation of generic computer components (weight noise supplying [unit]), mentally applying a noise by reflecting the noise to the initial weight (such as by adding the noise value to the initial weight value) to perform the operation using the new weight.
“a step of allowing the accuracy comparing unit to compare a first accuracy for a value calculated in the non-volatile memory array and a second accuracy for a value calculated in the robust memory array”; This limitation encompasses, excepting the recitation of generic computer components (the accuracy comparing unit), mentally comparing a first accuracy for a value calculated in the non-volatile memory array and a second accuracy for a value calculated in the robust memory array.
“a step of allowing the weight calibrator to perform backward propagation based on a comparison result performed in the accuracy comparing unit”; This limitation encompasses, excepting the recitation of generic computer components (the weight calibrator), a mathematical calculation of performing backward propagation based on the comparison result performed in the accuracy comparing unit.
Step 2A Prong 2: The judicial exception is not integrated into a practical application. The claim further recites that the method is performed in “a training operation variation compensating apparatus of an artificial neural network including a non-volatile memory array, a robust memory array, a weight noise supplying unit, an accuracy comparing unit, a weight copy engine, and a weight calibrator”, and that the noise is applied “to any one of the non-volatile memory array or the robust memory array,” however these limitations amount to mere instructions to apply an exception using a generic computer programmed with a generic class of computer algorithms (MPEP 2106.05(f)). The claim additionally recites that the operation is for training an artificial neural network, however this limitation amounts to merely generally linking the use of a judicial exception to the field of use of neural network training (MPEP 2106.05(h)).
Step 2B: The claim does not contain significantly more than the judicial exception. The analysis at this step mirrors that of Step 2A Prong 2 above. As an ordered whole, the claim is directed to a mentally performable process of performing an operation using a predetermined initial weight, applying noise to the initial weight to perform the operation using a weight obtained by reflecting the noise to the initial weight, comparing first and second accuracies, and performing backward propagation based on the result of the comparison. Nothing in the claim provides significantly more than this. As such, the claim is not patent eligible.
Claim 10
Step 1: A process, as above.
Step 2A Prong 1: The claim recites, inter alia:
“a step of allowing a weight copy engine to perform weight copy… based on the comparison result…”; This limitation encompasses, excepting the recitation of generic computer components (a weight copy engine), mentally performing weight copy based on the comparison result.
Step 2A Prong 2: The judicial exception is not integrated into a practical application. The claim further recites that the weight copy is performed by “a weight copy engine” and “from a memory array having a higher accuracy to a memory array having a lower accuracy,” however these limitations amount to mere instructions to apply an exception using a generic computer programmed with a generic class of computer algorithms (MPEP 2106.05(f)). The claim additionally recites that the comparison result is “obtained from the accuracy comparing unit, by the weight copying unit,” however this limitation amounts to the insignificant extra-solution activity of mere data gathering and outputting (MPEP 2106.05(g)).
Step 2B: The claim does not contain significantly more than the judicial exception. The “comparison result obtained from the accuracy comparing unit, by the weight copying unit” limitation, in addition to being insignificant extra solution activity, is also directed to the well-understood, routine, and conventional activity of receiving or transmitting data over a network (MPEP 2106.05(d)(II)(i) buySAFE, Inc. v. Google, Inc., 765 F.3d 1350, 1355, 112 USPQ2d 1093, 1096 (Fed. Cir. 2014) (computer receives and sends information over a network)). The further analysis at this step mirrors that of Step 2A Prong 2 above.
Claim 11
Step 1: A process, as above.
Step 2A Prong 1: The claim recites, inter alia:
“wherein in the step of allowing weight noise supplying to apply a noise, a noise is applied…at every predetermined first epoch to perform the… operation using the weight in which the noise is reflected”; This limitation encompasses, excepting the recitation of generic computer components (weight noise supplying), mentally applying a noise at every predetermined first epoch by reflecting the noise to a weight (such as by adding the noise value to the weight value) to perform the operation using the new weight.
Step 2A Prong 2: The judicial exception is not integrated into a practical application. The claim further recites that “weight noise supplying” applies a noise, and that the noise is applied “to the robust memory array,” however these limitations amount to mere instructions to apply an exception using a generic computer programmed with a generic class of computer algorithms (MPEP 2106.05(f)). The claim additionally recites that the operation is a “training” operation, however this limitation amounts to merely generally linking the use of a judicial exception to the field of use of neural network training (MPEP 2106.05(h)).
Step 2B: The claim does not contain significantly more than the judicial exception. The analysis at this step mirrors that of Step 2A Prong 2 above.
Claim 12
Step 1: A process, as above.
Step 2A Prong 1: The claim recites, inter alia:
“wherein when the first accuracy is lower than the second accuracy, if the weight copy engine performs the weight copy, the weight is copied…”; This limitation encompasses, excepting the recitation of generic computer components (the weight copy engine), mentally copying the weight when the first accuracy is lower than the second accuracy.
“when the weight calibrator performs the backward propagation based on the comparison result performed in the accuracy comparing unit, the weight calibrator performs the backward propagation based on a training operation performed in the robust memory array”; This limitation encompasses, excepting the recitation of generic computer components (the weight calibrator), a mathematical calculation of performing backward propagation based on a training operation performed in the robust memory array.
“when the first accuracy is equal to or higher than the second accuracy, the weight copy engine does not copy the weight and the weight calibrator does not perform the backward propagation”; This limitation encompasses, excepting the recitation of generic computer components (the weight copy engine and the weight calibrator), mentally deciding not to copy the weight and not to perform the backward propagation.
Step 2A Prong 2: The judicial exception is not integrated into a practical application. The claim further recites that the steps are performed by “the weight copy engine” and “the weight calibrator,” and the weight is copied “from the robust memory array to the non-volatile memory,” however these limitations amount to mere instructions to apply an exception using a generic computer programmed with a generic class of computer algorithms (MPEP 2106.05(f)).
Step 2B: The claim does not contain significantly more than the judicial exception. The analysis at this step mirrors that of Step 2A Prong 2 above.
Claim 13
Step 1: A process, as above.
Step 2A Prong 1: The claim recites, inter alia:
“wherein when the accuracy comparing unit compares the first accuracy and the second accuracy, the first accuracy and the second accuracy are compared at every predetermined second reference epoch”; This limitation encompasses, excepting the recitation of generic computer components (the accuracy comparing unit), mentally comparing the first and second accuracy at every predetermined second reference epoch.
“when the weight calibrator performs the backward propagation based on the comparison result performed in the accuracy comparing unit, if the first accuracy is lower than the second accuracy a predetermined reference epoch or more, the weight calibrator applies an calibration weight… to perform the training operation using the calibration weight”; This limitation encompasses, excepting the recitation of generic computer components (the weight calibrator), mentally applying a calibration weight to perform the training operation using the calibration weight.
Step 2A Prong 2: The judicial exception is not integrated into a practical application. The claim further recites that the steps are performed by “the accuracy comparing unit” and “the weight calibrator,” and the calibration weight is applied “to allow the non-volatile memory array to perform the training operation,” however these limitations amount to mere instructions to apply an exception using a generic computer programmed with a generic class of computer algorithms (MPEP 2106.05(f)).
Step 2B: The claim does not contain significantly more than the judicial exception. The analysis at this step mirrors that of Step 2A Prong 2 above.
Claim 14
Step 1: A method, as above.
Step 2A Prong 1: The claim recites the same judicial exception as claim 1.
Step 2A Prong 2: The judicial exception is not integrated into a practical application. The claim further recites “wherein when the non-volatile memory array performs the operation for training the artificial neural network using a predetermined initial weight, the non-volatile memory array is based on at least one of a magnetoresistive random access memory (MRAM), a phase change memory, and a ferroelectric random access memory (FeRAM) and when the robust memory array performs an operation for training the artificial neural network like the non-volatile memory array, the robust memory array is based on a static random access memory (SRAM),” however this limitation amounts to mere instructions to apply an exception using a generic computer programmed with a generic class of computer algorithms (MPEP 2106.05(f)).
Step 2B: The claim does not contain significantly more than the judicial exception. The analysis at this step mirrors that of Step 2A Prong 2 above.
Claim 15
Step 1: The claim is directed to non-statutory subject matter, however, for the purpose of the abstract idea rejection, Examiner will assume it is directed to the statutory category of articles of manufacture.
Step 2A Prong 1: The claim recites the same judicial exception as claim 9.
Step 2A Prong 2: The judicial exception is not integrated into a practical application. The claim further recites “A computer program stored in a computer readable recording medium to allow a computer to execute the training operation variation compensating method…” however, this limitation amounts to mere instructions to apply an exception using a generic computer programmed with a generic class of computer algorithms (MPEP 2106.05(f)). Otherwise, the analysis at this step mirrors that of claim 9.
Step 2B: The claim does not contain significantly more than the judicial exception. The analysis at this step mirrors that of claim 9.
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.
Claim Rejections - 35 USC § 103
Claims 1-15 are rejected under 35 U.S.C. 103 as being unpatentable over Ramesh et al. (US20220215235) (hereinafter “Ramesh”) in view of Vivekraja et al. (US12518167) (hereinafter “Vivekraja”), Seo et al. (US20240095528) (hereinafter “Seo”), and Xu et al. (US20220292401) (hereinafter “Xu”).
Regarding claim 1, Ramesh discloses “A training operation variation compensating apparatus of an artificial neural network, comprising:
a non-volatile memory array which performs an operation for training the artificial neural network using a predetermined initial weight (Ramesh, [0025]: “The memory system 104 can include one or more memory devices 126-1 to 126-N, which can include volatile and/or non-volatile memory cells” and [0037-0038]: “The memory system 204 can be analogous to the memory system 104 illustrated in FIG. 1, while the control circuitry 220 can be analogous to the control circuitry 120 illustrated in FIG. 1. As discussed above, the control circuitry 220 can control writing of the neural network 225 (e.g., an untrained neural network or partially trained neural network) to at least one of the memory devices 226-1 to 226-N and [0041]: “That is, in some embodiments, at least a portion of the operation to train the neural network 225 within the memory device 226-1 to can be performed while the neural network 225 is stored within the memory device 226-1 based on characteristics of the memory device 226-1 such as the type of media employed by the memory device 226-1, the bandwidth of the memory device 226-1, and/or the speed of the memory device 226-1, among others. In some embodiments, once the operation(s) to train the neural network 225 have been initiated, training operations can be performed within the memory device 226-1 in the absence of additional commands from the control circuitry 220 and/or a host (e.g., the host 102 illustrated in FIG. 1, herein)” and [0014]: “In one example, a neural network can be initialized with random weights”; Examiner notes that memory device 226-1 corresponds to “a non-volatile memory array” when it consists of non-volatile memory cells, and a first training operation being performed within memory device 226-1 in the absence of additional commands from the control circuitry and/or host, using a weight of the initialized weights, corresponds to “a non-volatile memory array which performs an operation for training the artificial neural network using a predetermined initial weight);
a robust memory array which performs an operation for training an artificial neural network… (Ramesh, [0025]: “The memory system 104 can include one or more memory devices 126-1 to 126-N, which can include volatile and/or non-volatile memory cells” and [0037-0038]: “The memory system 204 can be analogous to the memory system 104 illustrated in FIG. 1, while the control circuitry 220 can be analogous to the control circuitry 120 illustrated in FIG. 1. As discussed above, the control circuitry 220 can control writing of the neural network 225 (e.g., an untrained neural network or partially trained neural network) to at least one of the memory devices 226-1 to 226-N” and [0055]: “That is, in some embodiments, at least a portion of the operation to train the neural network 225 within the memory device 226-N to can be performed while the neural network 225 is stored within the memory device 226-N based on characteristics of the memory device 226-N such as the type of media employed by the memory device 226-N, the bandwidth of the memory device 226-N, and/or the speed of the memory device 226-N, among others. In some embodiments, once the operation(s) to train the neural network 225 have been initiated, training operations can be performed within the memory device 226-N in the absence of additional commands from the control circuitry 220 and/or a host (e.g., the host 102 illustrated in FIG. 1, herein)”; Examiner notes that memory device 226-N corresponds to “a robust memory array” and a first training operation being performed within memory device 226-N in the absence of additional commands from the control circuitry and/or host, corresponds to “a robust memory array which performs an operation for training an artificial neural network”).
Ramesh does not appear to explicitly disclose that the operation performed by the robust memory array is “the same as the non-volatile memory array using the initial weight,” or the further limitations of the claim.
However, Vivekraja discloses “a… [first worker node] which performs an operation for training… [an] artificial neural network using a predetermined initial weight (Vivekraja, Col 15, lines 45-63: “One way to accelerate a training process is by using a distributed system, to distribute the training process across multiple computing systems, each of which is configured as a worker node… A parallel training process (e.g., training processes 330a, 330b, 330p, etc.) can be performed at each worker node. For example, at around the same time, each worker node can perform operation 312 (e.g., 312a, 312b, 312p, etc.), which includes forward propagation operations, based on the respective sub-portion of training input data and the same set of initial weights”; Examiner notes that a first worker node of the multiple worker nodes corresponds to “a first worker node”, operation 312 corresponds to “an operation for training an artificial neural network,” and a weight of the set of initial weights corresponds to “a predetermined initial weight”);
a… [second worker node] which performs an operation for training an artificial neural network which is the same as the… [first worker node] using the initial weight” (Vivekraja, Col 15, lines 45-63: “One way to accelerate a training process is by using a distributed system, to distribute the training process across multiple computing systems, each of which is configured as a worker node… A parallel training process (e.g., training processes 330a, 330b, 330p, etc.) can be performed at each worker node. For example, at around the same time, each worker node can perform operation 312 (e.g., 312a, 312b, 312p, etc.), which includes forward propagation operations, based on the respective sub-portion of training input data and the same set of initial weights”; Examiner notes that a second worker node of the multiple worker nodes corresponds to “a second worker node”, operation 312 corresponds to “an operation for training an artificial neural network which is the same as the first worker node” and a weight of the set of initial weights corresponds to “the initial weight”).
Vivekraja and the instant application both relate to neural network training and are analogous. It would have been obvious to one of ordinary skill in the art, prior to the effective filing date of the claimed invention, to have modified Ramesh with the teachings of Vivekraja such that the operation performed by the robust memory array is “the same as the non-volatile memory using the initial weight,” and one would have been motivated to do so for the purpose of accelerating a training process (see Vivekraja, Col 16, lines 21-25).
Neither Ramesh nor Vivekraja appear to explicitly disclose the further limitations of the claim. However, Seo et al. discloses “a weight noise supplying [unit] (Seo, [0049]: “In some aspects of the present invention, software executing the instructions provided herein may be stored on a non-transitory computer-readable medium, wherein the software performs some or all of the steps of the present invention when executed on a processor”; Examiner notes that the software that performs the weight noise injection step corresponds to “a weight noise supplying [unit]”) which applies a noise to… [a] non-volatile memory array… to perform…[a] training operation using a weight obtained by reflecting the noise to… [an] initial weight” (Seo, [0079-0080]: “Now referring to FIG. 7, a diagram representing the overall method for machine learning in a temperature-resilient neural network model architecture is shown. At step 701, the method comprises loading a deep neural network model into a nonvolatile memory…Continuing to refer to FIG. 7, at step 702, the method further comprises training the deep neural network model using a progressive knowledge distillation algorithm as a function of a teacher model. A progressive knowledge distillation algorithm may be implemented at a thermal variation between 25 and 35 degrees Celsius with 20 epoch fine-tuning. As explained above, the algorithm comprises injecting, at step 703, using a clean model as the teacher model, low-temperature noise into a student model” and [0082]: “As shown in FIG. 6, distilling the knowledge in a step-by-step fashion in phase 1 enables the student model to learn the high-temperature variations while matching up with the teacher model that was trained with the low-temperature variations. To further improve the model's generality, the injected noises for each step, or temperature, are generated based on the temporally averaged variation between 0 and 10,000 seconds. The device noise is defined as the deviation between the programmed conductance and drifted conductance. Such bit-level noises are first transformed into the low-precision weight level distortions, as shown in FIG. 4. Specifically, for each temperature, the injected noise is the temporally averaged distortion between 0 and 10,000 seconds. The resultant noises are injected to the corresponded weight level during the PKD-BNA training”; Examiner notes that injecting low temperature noise into a student model corresponds to applying noise to a non-volatile memory array since the model is loaded into a nonvolatile memory array prior to training, the noises are injected to the corresponding weight levels, and a training operation is performed using the noise injected weights of the student model (see also Fig. 6))…
and a weight calibrator (Seo, [0049]: “In some aspects of the present invention, software executing the instructions provided herein may be stored on a non-transitory computer-readable medium, wherein the software performs some or all of the steps of the present invention when executed on a processor”; Examiner notes that the software that performs the back propagation step corresponds to “a weight calibrator”) which performs backward propagation based on… [a loss function] (Seo, Fig.6: see “Loss” and “Back Propagation”).
Seo and the instant application both relate to neural network training and in-memory computing and are analogous. It would have been obvious to one of ordinary skill in the art, prior to the effective filing date of the claimed invention, to have modified the combination of Ramesh and Vivekraja with the teachings of Seo to include “a weight noise supplying which applies a noise to any one of the non-volatile memory array or the robust memory array to perform the training operation using a weight obtained by reflecting the noise to the initial weight” and “a weight calibrator which performs backward propagation based on… [a loss function],” and one would have been motivated to do so for the purpose of achieving high robustness with largely improved accuracy of a neural network against temperature variations over time (see Seo, [0062]).
Neither Ramesh nor Vivekraja nor Seo appear to explicitly disclose the further limitations of the claim.
However, Xu discloses “an accuracy comparing unit which compares a first accuracy for a value calculated in… [a current student model] and a second accuracy for a value calculated in… [a candidate student model] (Xu, [0090]: “Upon updating teacher model 301, in step 610, a determination is made by decision maker 306 as to whether candidate student model 303 generates a better prediction of the observed target than current student model 302. Such a determination is based on how close the prediction is to the observed target”; Examiner notes that decision maker 306 corresponds to “an accuracy comparing unit,” how close the prediction generated by current student model 302 is to the observed target corresponds to “a first accuracy,” and how close the prediction generated by candidate student model 303 is to the observed target corresponds to “a second accuracy”);
and… performs… [a weight update] based on the comparison result performed in the accuracy comparing unit” (Xu, [0092]: “Furthermore, if candidate student model 303 is better at predicting the observed target than current student model 302, then in step 612, the current weights are updated with the new weights (new weights generated by teacher model 301 in step 604)”).
Xu and the instant application both relate to neural networks and are analogous. It would have been obvious to one of ordinary skill in the art, prior to the effective filing date of the claimed invention, to have modified the combination of Ramesh/Vivekraja/Seo with the teachings of Xu such that the apparatus includes an accuracy comparing unit which compares a first accuracy for a value calculated in the non-volatile memory array and a second accuracy for a value calculated in the robust memory array, and such that the weight calibrator performs the backward propagation based on the comparison result performed in the accuracy comparing unit, and one would have been motivated to do so for the purpose of identifying optimal weights to improve prediction accuracy (see Xu, [0001]).
Regarding claim 2, the rejection of claim 1 is incorporated. Ramesh as modified by Vivekraja, Seo, and Xu, further discloses “a… copy engine which performs… [neural network data] copy from a [first] memory array to a [second] memory array” (Ramesh, [0064]: “In this example, the control circuitry 220 can write data corresponding to a neural network 225 to a first memory device (e.g., the memory device 226-1) among the plurality of memory devices 226-1 to 226-N and cause, while the neural network 225 is stored in the first memory device 226-1, at least a first portion of a training operation for a neural network 225 by determining one or more first weights for a hidden layer of the neural network 225 to be performed. The control circuitry 220 can then write the data corresponding to the neural network 225 to a second memory device (e.g., the memory device 226-N) and cause, while the neural network 225 is stored in the second memory device, at least a second portion of the training operation for the neural network 225 by determining one or more second weights for the hidden layer of the neural network 225 to be performed”; Examiner notes control circuitry 220 corresponds to “a copy engine,” which copies the neural network from a first memory array 226-1 to a second memory array 226-N).
Neither Ramesh nor Vivekraja nor Seo appear to explicitly disclose the further limitations of the claim.
However, Xu discloses “a weight copy engine which performs weight copy from a… [model] having a higher accuracy to a… [model] having lower accuracy based on the comparison result obtained from the accuracy comparing unit, by the weight copying unit” ([0090]: “Upon updating teacher model 301, in step 610, a determination is made by decision maker 306 as to whether candidate student model 303 generates a better prediction of the observed target than current student model 302. Such a determination is based on how close the prediction is to the observed target” and [0092]: “Furthermore, if candidate student model 303 is better at predicting the observed target than current student model 302, then in step 612, the current weights are updated with the new weights (new weights generated by teacher model 301 in step 604)”; Examiner notes that the current weights which were used to train the lower accuracy model “current student model 302” are updated with the new weights used to train the higher accuracy model “candidate student model 303” (see also Fig. 3), and the software that performs step 612 corresponds to “a weight copy engine/weight copying unit” (see [0041]).
Xu and the instant application both relate to neural networks and are analogous. It would have been obvious to one of ordinary skill in the art, prior to the effective filing date of the claimed invention, to have modified the combination of Ramesh/Vivekraja/Seo with the teachings of Xu such that the apparatus further comprises “a weight copy engine which performs weight copy from a memory array having a higher accuracy to a memory array having a lower accuracy based on the comparison result obtained from the accuracy comparing unit, by the weight copying unit,” and one would have been motivated to do so for the purpose of identifying optimal weights to improve prediction accuracy (see Xu, [0001]).
Regarding claim 3, the rejection of claim 2 is incorporated. Neither Ramesh nor Vivekraja nor Xu appear to explicitly disclose the further limitations of the claim.
However, Seo further discloses “wherein the weight noise supplying unit applies a noise to… [a] memory array at every predetermined first epoch to perform the training operation using the weight in which the noise is reflected” (Seo, Fig. 6: Examiner notes that the noise is injected at least once every 20 epochs as shown in Fig. 6, therefore the first epoch of every 20-epoch set corresponds to “every predetermined first epoch”).
Seo and the instant application both relate to neural network training and in-memory computing and are analogous. It would have been obvious to one of ordinary skill in the art, prior to the effective filing date of the claimed invention, to have modified the combination of Ramesh/Vivekraja/Xu with the teachings of Seo such that “the weight noise supplying unit applies a noise to the robust memory array at every predetermined first epoch to perform the training operation using the weight in which the noise is reflected,” and one would have been motivated to do so for the purpose of achieving high robustness with largely improved accuracy of a neural network against temperature variations over time (see Seo, [0062]).
Regarding claim 4, the rejection of claim 3 is incorporated. Ramesh as modified by Vivekraja, Seo, and Xu discloses a non-volatile memory array having a first accuracy, a robust memory array having a second accuracy (see rejection of claim 1), and “a weight copy engine which performs weight copy from a memory array having a higher accuracy to a memory array having a lower accuracy” (see rejection of claim 2), but does not appear to explicitly disclose the further limitations of the claim.
However, Seo further discloses “…the weight calibrator performs the backward propagation based on a training operation performed in… [a] memory array” (Seo, Fig. 6: Examiner notes that Back Propagation after forward pass through the model loaded in a memory array corresponds to performing the backward propagation based on a training operation performed in a memory array).
Seo and the instant application both relate to neural network training and in-memory computing and are analogous. It would have been obvious to one of ordinary skill in the art, prior to the effective filing date of the claimed invention, to have modified the combination of Ramesh/Vivekraja/Xu with the teachings of Seo such that “the weight calibrator performs the backward propagation based on a training operation performed in the robust memory array,” and one would have been motivated to do so for the purpose of achieving high robustness with largely improved accuracy of a neural network against temperature variations over time (see Seo, [0062]).
Neither Ramesh nor Vivekraja nor Seo appear to explicitly disclose the further limitations of the claim.
However, Xu discloses “wherein when the first accuracy is lower than the second accuracy, the weight copy engine performs the weight copy from… [a higher accuracy model] to… [a lower accuracy model] and… [a model update unit performs updating the lower accuracy model with the higher accuracy model]” (Xu, [0091]: “If candidate student model 303 is better at predicting the observed target than current student model 302, then, in step 611, current student model 302 is updated with candidate student model 303. That is, in step 611, current student model 302 is essentially replaced with candidate student model 303” and Xu, [0092]: “Furthermore, if candidate student model 303 is better at predicting the observed target than current student model 302, then in step 612, the current weights are updated with the new weights (new weights generated by teacher model 301 in step 604)”; Examiner notes that the candidate student model corresponds to a higher accuracy model, the current student model corresponds to a lower accuracy model, and the software that performs step 611 corresponds to “a model update unit” (see [0041])).
Xu and the instant application both relate to neural networks and are analogous. It would have been obvious to one of ordinary skill in the art, prior to the effective filing date of the claimed invention, to have modified the combination of Ramesh/Vivekraja/Seo with the teachings of Xu such that “when the first accuracy is lower than the second accuracy, the weight copy engine performs the weight copy from the robust memory array to the non-volatile memory array and the weight calibrator performs the backward propagation based on a training operation performed in the robust memory array,” and one would have been motivated to do so for the purpose of updating the model in the lower accuracy memory array (non-volatile memory) with the model in the higher accuracy memory array (robust memory) to identify optimal weights to improve prediction accuracy (see Xu, [0001]).
Regarding claim 5, the rejection of claim 3 is incorporated. Neither Ramesh nor Vivekraja nor Seo appear to explicitly disclose the further limitations of the claim.
However, Xu further discloses “wherein when the first accuracy is equal to or higher than the second accuracy, the weight copy engine does not copy the weight and… [a model update unit does not perform a model update]” (Xu, [0094]: “Alternatively, if candidate student model 303 is not better at predicting the observed target than current student model 302, then decision maker 306 directly requests the updated teacher model 301 (updated in step 609) to generate new weights in step 604 using the current student features from the current student model 302 and current weights” and Xu, Fig. 6: if candidate student model is not better, (NO arrow), steps 611 and 612 are skipped; Examiner notes that the software that performs step 611 corresponds to “a model update unit” (see [0041])).
Xu and the instant application both relate to neural networks and are analogous. It would have been obvious to one of ordinary skill in the art, prior to the effective filing date of the claimed invention, to have modified the combination of Ramesh/Vivekraja/Seo with the teachings of Xu such that “when the first accuracy is equal to or higher than the second accuracy, the weight copy engine does not copy the weight and the weight calibrator does not performs the backward propagation” and one would have been motivated to do so for the purpose of identifying optimal weights to improve prediction accuracy in machine learning techniques (see Xu, [0001]).
Regarding claim 6, the rejection of claim 1 is incorporated. Neither Ramesh nor Vivekraja nor Seo appear to explicitly disclose the further limitations of the claim.
However, Xu further discloses “wherein the accuracy comparing unit compares the first accuracy and the second accuracy at every predetermined second reference epoch (Xu, Fig 6: Examiner notes accuracy comparing step 610 occurs each time step 605 of training a candidate student model using training data occurs (corresponding to an epoch because it is a pass of the training data through the model), therefore each epoch can be considered a predetermined second reference epoch) and when the first accuracy is lower than the second accuracy a predetermined reference epoch or more… [a teacher model] applies an calibration weight to allow the… [candidate student model] to perform the training operation using the calibration weight (Xu, Fig 6: When candidate student model is not better (NO arrow), teacher model generates new weights (604) which are used by the candidate student model during a training operation (605)).
Xu and the instant application both relate to neural networks and are analogous. It would have been obvious to one of ordinary skill in the art, prior to the effective filing date of the claimed invention, to have modified the combination of Ramesh/Vivekraja/Seo with the teachings of Xu such that the accuracy comparing unit compares the first accuracy and the second accuracy at every predetermined second reference epoch, and when the first accuracy is lower than the second accuracy a predetermined reference epoch or more, the weight calibrator applies an calibration weight to allow the non-volatile memory array to perform the training operation using the calibration weight, and one would have been motivated to do so for the purpose of identifying optimal weights to improve prediction accuracy in machine learning techniques (see Xu, [0001]).
Regarding claim 7, the rejection of claim 1 is incorporated. Ramesh as modified by Vivekraja, Seo, and Xu further discloses “wherein the non-volatile memory array is based on at least one of a magnetoresistive random access memory (MRAM), a phase change memory, and a ferroelectric random access memory (FeRAM)” (Ramesh, [0026]: “Embodiments are not so limited, however, and the memory system 104 can include other non-volatile memory devices 126-1 to 126-N such as non-volatile random-access memory devices (e.g., NVRAM, ReRAM, FeRAM, MRAM, PCM)”).
Regarding claim 8, the rejection of claim 1 is incorporated. Ramesh as modified by Vivekraja, Seo, and Xu further discloses “wherein the robust memory array is based on a static random access memory (SRAM)” (Ramesh, [0002]: “Volatile memory can require power to maintain its data (e.g., host data, error data, etc.) and includes random access memory (RAM), dynamic random access memory (DRAM), static random access memory (SRAM), synchronous dynamic random access memory (SDRAM), and thyristor random access memory (TRAM), among others”; Examiner notes that the robust memory array can consists of volatile memory cells (see rejection of claim 1), and thus can be based on SRAM).
Regarding claim 9, Ramesh discloses “A training operation variation compensating apparatus of an artificial neural network including a non-volatile memory array, a robust memory array, a weight noise supplying unit, an accuracy comparing unit, a weight copy engine, and a weight calibrator, comprising:
a step of allowing the non-volatile memory array… to perform an operation for training an artificial neural network using a predetermined initial weight… (Ramesh, [0025]: “The memory system 104 can include one or more memory devices 126-1 to 126-N, which can include volatile and/or non-volatile memory cells” and [0037-0038]: “The memory system 204 can be analogous to the memory system 104 illustrated in FIG. 1, while the control circuitry 220 can be analogous to the control circuitry 120 illustrated in FIG. 1. As discussed above, the control circuitry 220 can control writing of the neural network 225 (e.g., an untrained neural network or partially trained neural network) to at least one of the memory devices 226-1 to 226-N and [0041]: “That is, in some embodiments, at least a portion of the operation to train the neural network 225 within the memory device 226-1 to can be performed while the neural network 225 is stored within the memory device 226-1 based on characteristics of the memory device 226-1 such as the type of media employed by the memory device 226-1, the bandwidth of the memory device 226-1, and/or the speed of the memory device 226-1, among others. In some embodiments, once the operation(s) to train the neural network 225 have been initiated, training operations can be performed within the memory device 226-1 in the absence of additional commands from the control circuitry 220 and/or a host (e.g., the host 102 illustrated in FIG. 1, herein)” and [0014]: “In one example, a neural network can be initialized with random weights”; Examiner notes that memory device 226-1 corresponds to “a non-volatile memory array” when it consists of non-volatile memory cells, and a first training operation being performed within memory device 226-1 in the absence of additional commands from the control circuitry and/or host, using a weight of the initialized weights, corresponds to “a non-volatile memory array which performs an operation for training the artificial neural network using a predetermined initial weight);
[a step of allowing] the robust memory array [to perform an operation for training an artificial neural network]…
(Ramesh, [0025]: “The memory system 104 can include one or more memory devices 126-1 to 126-N, which can include volatile and/or non-volatile memory cells” and [0037-0038]: “The memory system 204 can be analogous to the memory system 104 illustrated in FIG. 1, while the control circuitry 220 can be analogous to the control circuitry 120 illustrated in FIG. 1. As discussed above, the control circuitry 220 can control writing of the neural network 225 (e.g., an untrained neural network or partially trained neural network) to at least one of the memory devices 226-1 to 226-N” and [0055]: “That is, in some embodiments, at least a portion of the operation to train the neural network 225 within the memory device 226-N to can be performed while the neural network 225 is stored within the memory device 226-N based on characteristics of the memory device 226-N such as the type of media employed by the memory device 226-N, the bandwidth of the memory device 226-N, and/or the speed of the memory device 226-N, among others. In some embodiments, once the operation(s) to train the neural network 225 have been initiated, training operations can be performed within the memory device 226-N in the absence of additional commands from the control circuitry 220 and/or a host (e.g., the host 102 illustrated in FIG. 1, herein)”; Examiner notes that memory device 226-N corresponds to “a robust memory array” and a first training operation being performed within memory device 226-N in the absence of additional commands from the control circuitry and/or host, corresponds to “a robust memory array which performs an operation for training an artificial neural network”).
Ramesh does not appear to explicitly disclose that the operation performed by the robust memory array is the same as the non-volatile memory array using the same predetermined initial weight, or the further limitations of the claim.
However, Vivekraja discloses “a step of allowing… [a first worker node and a second worker node] to perform an operation for training an artificial neural network using a predetermined initial weight” (Vivekraja, Col 15, lines 45-63: “One way to accelerate a training process is by using a distributed system, to distribute the training process across multiple computing systems, each of which is configured as a worker node… A parallel training process (e.g., training processes 330a, 330b, 330p, etc.) can be performed at each worker node. For example, at around the same time, each worker node can perform operation 312 (e.g., 312a, 312b, 312p, etc.), which includes forward propagation operations, based on the respective sub-portion of training input data and the same set of initial weights”; Examiner notes that a first worker node of the multiple worker nodes corresponds to “a first worker node”, a second worker node of the multiple worker nodes corresponds to “a second worker node”, operation 312 corresponds to “an operation for training an artificial neural network,” and a weight of the set of initial weights corresponds to “a predetermined initial weight”).
Vivekraja and the instant application both relate to neural network training and are analogous. It would have been obvious to one of ordinary skill in the art, prior to the effective filing date of the claimed invention, to have modified Ramesh with the teachings of Vivekraja such that the non-volatile memory array and the robust memory array both “perform an operation for training an artificial neural network using a predetermined initial weight,” and one would have been motivated to do so for the purpose of accelerating a training process (see Vivekraja, Col 16, lines 21-25).
Neither Ramesh nor Vivekraja appear to explicitly disclose the further limitations of the claim. However, Seo et al. discloses “a step of allowing weight noise supplying [unit] (Seo, [0049]: “In some aspects of the present invention, software executing the instructions provided herein may be stored on a non-transitory computer-readable medium, wherein the software performs some or all of the steps of the present invention when executed on a processor”; Examiner notes that the software that performs the weight noise injection step corresponds to “a weight noise supplying [unit]”) to apply a noise to… [a] non-volatile memory array… to perform…[a] training operation using a weight obtained by reflecting the noise to… [an] initial weight” (Seo, [0079-0080]: “Now referring to FIG. 7, a diagram representing the overall method for machine learning in a temperature-resilient neural network model architecture is shown. At step 701, the method comprises loading a deep neural network model into a nonvolatile memory…Continuing to refer to FIG. 7, at step 702, the method further comprises training the deep neural network model using a progressive knowledge distillation algorithm as a function of a teacher model. A progressive knowledge distillation algorithm may be implemented at a thermal variation between 25 and 35 degrees Celsius with 20 epoch fine-tuning. As explained above, the algorithm comprises injecting, at step 703, using a clean model as the teacher model, low-temperature noise into a student model” and [0082]: “As shown in FIG. 6, distilling the knowledge in a step-by-step fashion in phase 1 enables the student model to learn the high-temperature variations while matching up with the teacher model that was trained with the low-temperature variations. To further improve the model's generality, the injected noises for each step, or temperature, are generated based on the temporally averaged variation between 0 and 10,000 seconds. The device noise is defined as the deviation between the programmed conductance and drifted conductance. Such bit-level noises are first transformed into the low-precision weight level distortions, as shown in FIG. 4. Specifically, for each temperature, the injected noise is the temporally averaged distortion between 0 and 10,000 seconds. The resultant noises are injected to the corresponded weight level during the PKD-BNA training”; Examiner notes that injecting low temperature noise into a student model corresponds to applying noise to a non-volatile memory array since the model is loaded into a nonvolatile memory array prior to training, the noises are injected to the corresponding weight levels, and a training operation is performed using the noise injected weights of the student model (see also Fig. 6))…
and a step of allowing the weight calibrator (Seo, [0049]: “In some aspects of the present invention, software executing the instructions provided herein may be stored on a non-transitory computer-readable medium, wherein the software performs some or all of the steps of the present invention when executed on a processor”; Examiner notes that the software that performs the back propagation step corresponds to “a weight calibrator”) to perform backward propagation based on… [a loss function] (Seo, Fig.6: see “Loss” and “Back Propagation”).
Seo and the instant application both relate to neural network training and in-memory computing and are analogous. It would have been obvious to one of ordinary skill in the art, prior to the effective filing date of the claimed invention, to have modified the combination of Ramesh and Vivekraja with the teachings of Seo to include “a step of allowing weight noise supplying to apply a noise to any one of the non-volatile memory array or the robust memory array to perform the training operation using a weight obtained by reflecting the noise to the initial weight” and “a step of allowing the weight calibrator to perform backward propagation based on… [a loss function],” and one would have been motivated to do so for the purpose of achieving high robustness with largely improved accuracy of a neural network against temperature variations over time (see Seo, [0062]).
Neither Ramesh nor Vivekraja nor Seo appear to explicitly disclose the further limitations of the claim.
However, Xu discloses “a step of allowing the accuracy comparing unit to compare a first accuracy for a value calculated in… [a candidate student model] and a second accuracy for a value calculated in… [a current student model] (Xu, [0090]: “Upon updating teacher model 301, in step 610, a determination is made by decision maker 306 as to whether candidate student model 303 generates a better prediction of the observed target than current student model 302. Such a determination is based on how close the prediction is to the observed target”; Examiner notes that decision maker 306 corresponds to “an accuracy comparing unit,” how close the prediction generated by current student model 302 is to the observed target corresponds to “a first accuracy,” and how close the prediction generated by candidate student model 303 is to the observed target corresponds to “a second accuracy”);
and… perform… [a weight update] based on the comparison result performed in the accuracy comparing unit” (Xu, [0092]: “Furthermore, if candidate student model 303 is better at predicting the observed target than current student model 302, then in step 612, the current weights are updated with the new weights (new weights generated by teacher model 301 in step 604)”).
Xu and the instant application both relate to neural networks and are analogous. It would have been obvious to one of ordinary skill in the art, prior to the effective filing date of the claimed invention, to have modified the combination of Ramesh/Vivekraja/Seo with the teachings of Xu such that the method includes “a step of allowing the accuracy comparing unit to compare a first accuracy for a value calculated in the non-volatile memory array and a second accuracy for a value calculated in the robust memory array,” and such that the weight calibrator performs the backward propagation based on a comparison result performed in the accuracy comparing unit, and one would have been motivated to do so for the purpose of identifying optimal weights to improve prediction accuracy (see Xu, [0001]).
Regarding claim 10, the rejection of claim 9 is incorporated. Ramesh as modified by Vivekraja, Seo, and Xu, further discloses “a step of allowing a… copy engine to perform… [neural network data] copy from a [first] memory array to a [second] memory array” (Ramesh, [0064]: “In this example, the control circuitry 220 can write data corresponding to a neural network 225 to a first memory device (e.g., the memory device 226-1) among the plurality of memory devices 226-1 to 226-N and cause, while the neural network 225 is stored in the first memory device 226-1, at least a first portion of a training operation for a neural network 225 by determining one or more first weights for a hidden layer of the neural network 225 to be performed. The control circuitry 220 can then write the data corresponding to the neural network 225 to a second memory device (e.g., the memory device 226-N) and cause, while the neural network 225 is stored in the second memory device, at least a second portion of the training operation for the neural network 225 by determining one or more second weights for the hidden layer of the neural network 225 to be performed”; Examiner notes control circuitry 220 corresponds to “a copy engine,” which copies the neural network from a first memory array 226-1 to a second memory array 226-N).
Neither Ramesh nor Vivekraja nor Seo appear to explicitly disclose the further limitations of the claim.
However, Xu discloses “a step of allowing a weight copy engine to performs weight copy from a… [model] having a higher accuracy to a… [model] having lower accuracy based on the comparison result obtained from the accuracy comparing unit, by the weight copying unit” ([0090]: “Upon updating teacher model 301, in step 610, a determination is made by decision maker 306 as to whether candidate student model 303 generates a better prediction of the observed target than current student model 302. Such a determination is based on how close the prediction is to the observed target” and [0092]: “Furthermore, if candidate student model 303 is better at predicting the observed target than current student model 302, then in step 612, the current weights are updated with the new weights (new weights generated by teacher model 301 in step 604)”; Examiner notes that the current weights which were used to train the lower accuracy model “current student model 302” are updated with the new weights used to train the higher accuracy model “candidate student model 303” (see also Fig. 3), and the software that performs step 612 corresponds to “a weight copy engine/copying unit” (see [0041]).
Xu and the instant application both relate to neural networks and are analogous. It would have been obvious to one of ordinary skill in the art, prior to the effective filing date of the claimed invention, to have modified the combination of Ramesh/Vivekraja/Seo with the teachings of Xu such that the apparatus further comprises “a weight copy engine which performs weight copy from a memory array having a higher accuracy to a memory array having a lower accuracy based on the comparison result obtained from the accuracy comparing unit, by the weight copying unit,” and one would have been motivated to do so for the purpose of identifying optimal weights to improve prediction accuracy (see Xu, [0001]).
Regarding claim 11, the rejection of claim 10 is incorporated. Neither Ramesh nor Vivekraja nor Xu appear to explicitly disclose the further limitations of the claim.
However, Seo further discloses “wherein in the step of allowing weight noise supplying to apply a noise, a noise is applied to… [a] memory array at every predetermined first epoch to perform the training operation using the weight in which the noise is reflected” (Seo, Fig. 6: Examiner notes that the noise is injected at least once every 20 epochs as shown in Fig. 6, therefore the first epoch of every 20-epoch set corresponds to “every predetermined first epoch”).
Seo and the instant application both relate to neural network training and in-memory computing and are analogous. It would have been obvious to one of ordinary skill in the art, prior to the effective filing date of the claimed invention, to have modified the combination of Ramesh/Vivekraja/Xu with the teachings of Seo such that “the weight noise supplying unit applies a noise to the robust memory array at every predetermined first epoch to perform the training operation using the weight in which the noise is reflected,” and one would have been motivated to do so for the purpose of achieving high robustness with largely improved accuracy of a neural network against temperature variations over time (see Seo, [0062]).
Regarding claim 12, the rejection of claim 11 is incorporated. Ramesh as modified by Vivekraja, Seo, and Xu discloses a non-volatile memory array having a first accuracy, a robust memory array having a second accuracy (see rejection of claim 9), and a weight copy engine which performs weight copy from a memory array having a higher accuracy to a memory array having a lower accuracy (see rejection of claim 10), but does not appear to explicitly disclose the further limitations of the claim.
However, Seo further discloses “…the weight calibrator performs the backward propagation based on a training operation performed in… [a] memory array” (Seo, Fig. 6: Examiner notes that Back Propagation after forward pass through the model loaded in a memory array corresponds to performing the backward propagation based on a training operation performed in a memory array).
Seo and the instant application both relate to neural network training and in-memory computing and are analogous. It would have been obvious to one of ordinary skill in the art, prior to the effective filing date of the claimed invention, to have modified the combination of Ramesh/Vivekraja/Xu with the teachings of Seo such that “when the weight calibrator performs the backward propagation based on the comparison result performed in the accuracy comparing unit, the weight calibrator performs the backward propagation based on a training operation performed in the robust memory array,” and one would have been motivated to do so for the purpose of achieving high robustness with largely improved accuracy of a neural network against temperature variations over time (see Seo, [0062]).
Neither Ramesh nor Vivekraja nor Seo appear to explicitly disclose the further limitations of the claim.
However, Xu discloses “wherein when the first accuracy is lower than the second accuracy, if the weight copy engine performs the weight copy, the weight is copied from… [a higher accuracy model] to… [a lower accuracy model] and …[a model update unit performs updating the lower accuracy model with the higher accuracy model] (Xu, [0091]: “If candidate student model 303 is better at predicting the observed target than current student model 302, then, in step 611, current student model 302 is updated with candidate student model 303. That is, in step 611, current student model 302 is essentially replaced with candidate student model 303” and Xu, [0092]: “Furthermore, if candidate student model 303 is better at predicting the observed target than current student model 302, then in step 612, the current weights are updated with the new weights (new weights generated by teacher model 301 in step 604)”; Examiner notes that the candidate student model corresponds to a higher accuracy model, the current student model corresponds to a lower accuracy model, and the software that performs step 611 corresponds to “a model update unit” (see [0041])) and
when the first accuracy is equal to or higher than the second accuracy, the weight copy engine does not copy the weight and… [the model update unit does not perform the model update]” (Xu, [0094]: “Alternatively, if candidate student model 303 is not better at predicting the observed target than current student model 302, then decision maker 306 directly requests the updated teacher model 301 (updated in step 609) to generate new weights in step 604 using the current student features from the current student model 302 and current weights” and Xu, Fig. 6: if candidate student model is not better, (NO arrow), steps 611 and 612 are skipped).
Xu and the instant application both relate to neural networks and are analogous. It would have been obvious to one of ordinary skill in the art, prior to the effective filing date of the claimed invention, to have modified the combination of Ramesh/Vivekraja/Seo with the teachings of Xu such that “when the first accuracy is lower than the second accuracy, if the weight copy engine performs the weight copy, the weight is copied from the robust memory array to the non-volatile memory and when the weight calibrator performs the backward propagation based on the comparison result performed in the accuracy comparing unit, the weight calibrator performs the backward propagation based on a training operation performed in the robust memory array, and when the first accuracy is equal to or higher than the second accuracy, the weight copy engine does not copy the weight and the weight calibrator does not perform the backward propagation,” and one would have been motivated to do so for the purpose of updating the model in the lower accuracy memory array (non-volatile memory) with the model in the higher accuracy memory array (robust memory) to identify optimal weights to improve prediction accuracy in machine learning techniques (see Xu, [0001]).
Regarding claim 13, the rejection of claim 9 is incorporated. Neither Ramesh nor Vivekraja nor Seo appear to explicitly disclose the further limitations of the claim.
However, Xu further discloses “wherein when the accuracy comparing unit compares the first accuracy and the second accuracy, the first accuracy and the second accuracy are compared at every predetermined second reference epoch (Xu, Fig 6: Examiner notes accuracy comparing step 610 occurs each time step 605 of training a candidate student model using training data occurs (corresponding to an epoch because it is a pass of the training data through the model), therefore each epoch can be considered a predetermined second reference epoch) and… if the first accuracy is lower than the second accuracy a predetermined reference epoch or more… [a teacher model] applies an calibration weight to allow the… [candidate student model] to perform the training operation using the calibration weight (Xu, Fig 6: Examiner notes when candidate student model is not better (NO arrow), teacher model generates new weights (604) which are used by the candidate student model during a training operation (605)).
Xu and the instant application both relate to neural networks and are analogous. It would have been obvious to one of ordinary skill in the art, prior to the effective filing date of the claimed invention, to have modified the combination of Ramesh/Vivekraja/Seo with the teachings of Xu such that “when the accuracy comparing unit compares the first accuracy and the second accuracy, the first accuracy and the second accuracy are compared at every predetermined second reference epoch and when the weight calibrator performs the backward propagation based on the comparison result performed in the accuracy comparing unit, if the first accuracy is lower than the second accuracy a predetermined reference epoch or more, the weight calibrator applies an calibration weight to allow the non-volatile memory array to perform the training operation using the calibration weight,” and one would have been motivated to do so for the purpose of improving prediction accuracy in machine learning techniques (see Xu, [0001]).
Regarding claim 14, the rejection of claim 9 is incorporated. Ramesh as modified by Vivekraja, Seo, and Xu further discloses “wherein when the non-volatile memory array performs the operation for training the artificial neural network using a predetermined initial weight, the non-volatile memory array is based on at least one of a magnetoresistive random access memory (MRAM), a phase change memory, and a ferroelectric random access memory (FeRAM) (Ramesh, [0026]: “Embodiments are not so limited, however, and the memory system 104 can include other non-volatile memory devices 126-1 to 126-N such as non-volatile random-access memory devices (e.g., NVRAM, ReRAM, FeRAM, MRAM, PCM)”) and when the robust memory array performs an operation for training the artificial neural network like the non-volatile memory array, the robust memory array is based on a static random access memory (SRAM)” (Ramesh, [0002]: “Volatile memory can require power to maintain its data (e.g., host data, error data, etc.) and includes random access memory (RAM), dynamic random access memory (DRAM), static random access memory (SRAM), synchronous dynamic random access memory (SDRAM), and thyristor random access memory (TRAM), among others”; Examiner notes that the robust memory array may consist of volatile memory cells (see rejection of claim 9), and thus can be based on SRAM).
Regarding claim 15, the rejection of claim 9 is incorporated. Ramesh as modified by Vivekraja, Seo, and Xu further discloses “A computer program stored in a computer readable recording medium to allow a computer to execute the training operation variation compensating method of an artificial neural network according to claim 9” (Ramesh, [0072]: “FIG. 4 is a flow diagram 430 corresponding to a memory system to train neural networks in accordance with a number of embodiments of the present disclosure. The flow 430 can be performed by processing logic that can include hardware (e.g., processing device(s), control circuitry, dedicated logic, programmable logic, microcode, hardware of a device, and/or integrated circuit(s), etc.), software (e.g., instructions run or executed on a processing device), or a combination thereof”).
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to GWYNEVERE A DETERDING whose telephone number is (571)272-7657. The examiner can normally be reached Mon-Fri. 9am-5pm.
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.
/G.A.D./Examiner, Art Unit 2125
/KAMRAN AFSHAR/Supervisory Patent Examiner, Art Unit 2125
Prosecution Insights