DETAILED ACTION
This Office Action is sent in response to the Applicant’s Communication received on 12/05/2025 for application number 17/824,832. The Office hereby acknowledges receipt of the following and placed of record in file: Specification, Drawings, Abstract, Oath/Declaration, IDS, and Claims.
Claims 1-30 are pending.
Claims 1, 4, 6, 8, 9, 12, 14, 16, 17, 20, 21, 23, 24, 27, 28, and 30 are currently amended.
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 .
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on 12/05/2025 has been entered.
Response to Argument
35 USC 101
On page 10 of the remarks section, Applicant argues that independent claim 1 recites elements such as "computing an approximation of a posterior probability using a variational distribution of weights based on the extracted set of features of the one or more personal examples, " and "generating, at the edge device, a personal model using the shared neural network model, the personal model including a set of personalized weights determined at run-time based on the approximated posterior probability, "
which as a whole provide an improvement to the technological fields of model personalization for deployment on resource-limited edge devices. As described in the present specification:
Personalization on the edge devices has become an important issue. Personalization involves tailoring a global model to the distribution of personal domains. Conventional approaches to personalization may involve retraining the global model with on-device learning at each personal device. However, such on-device learning is computationally expensive and time consuming on personal devices because of the limited resources.
One challenge for personalization is due to the difficulty in collecting enough personal data. Conventional approaches have attempted to localize multiple models to different personal domains, and then merge the localized model to obtain a universal model. However, the universal model is unable to completely cover a universal domain. In addition, the deployed universal model may encounter various unprecedented personal domains. Although studies in domain adaptation fields have attempted to transfer the knowledge from a source domain to a target domain, these approaches use large amounts of labeled or unlabeled data in the source domain. Therefore, it remains challenging to flexibly personalize a model to a given personal domain using only small-scaled personal data. See, published specification, paragraphs 25-26 (emphasis added).
Applicant further continues, to address the identified challenges, the elements of independent claim 1 beneficially enables model personalization with increased model accuracy while reducing computation and power consumption of resource-limited edge devices. As described, for instance, in the published specification:
Aspects of the present disclosure are directed to generating a personalized neural network model using few-shot personalization. Few shot personalization is a model personalization based on very few personal examples (e.g., examples that may not have been included in the training examples used in generating the
shared model). Given personal data, weights of layers specialized to its
personality may be computed on-the-fly (e.g., during runtime) via forwarding only a few personal data examples. A personality may refer to a data distribution of the personal domain on the edge device. A meta unit is trained to capture weight distribution of layers that fit a target person. For instance, the meta unit may be trained to produce an approximated posterior distribution of weights of a layer based on the personality. As the data in a personal domain share the same personality, the target prior distribution may, for example, be set as the averaged distribution of the data in a variational inference of the meta unit. Published specification, paragraph 59 (emphasis added).
Therefore, Applicant concludes that independent claim 1 is believed to be directed to patent eligible subject matter.
The Examiner respectfully disagrees. Although the instant application paragraph 59 has detailed descriptions of implementing steps that provide the technological improvement of enabling model personalization with increased model accuracy while reducing computation and power consumption of resource-limited edge devices, these detailed inventive steps are not explicitly present in the current claim. Specifically, “generating a personalized neural network model using few-shot personalization”, “weights of layers specialized to its personality”, personality referring to a data distribution of the personal domain on the edge device, “forwarding only a few personal data examples”, and “produce an approximated posterior distribution of weights of a layer based on the personality” are not explicitly stated in the claim. Therefore, the claimed elements or combination of elements do not reflect the technical steps leading to an improvement as alleged. If the specification sets forth an improvement in technology, the claim must be evaluated to ensure that the claim itself reflects the disclosed improvement. That is, the claim includes the components or steps of the invention that provide the improvement described in the specification.
Therefore, the 35 USC 101 rejection is maintained.
35 USC 103
Applicant further argues that independent claim 1 has been amended and recites inter alia "computing an approximation of a posterior probability using a variational distribution of weights based on the extracted set of features of the one or more personal examples." None of the applied references appear to disclose at least this portion of claim 1. Viswanathan appears to determine the model weights by applying a gradient function calculated using backpropagation. See, Viswanathan, paragraph 79. Huang describes an image segmentation model in which a target splicing vector is obtained and used to compute a first posterior probability that a row image includes text content. See, Huang, paragraph 45. Wen describes obtaining multiple machine learning models from a remote system and storing the models locally on a user computing device. The locally stored machine learning models are used multiple times without requesting the machine learning models again from the remote system. One of the machine learning models serves as a base model, and the other machine learning
models serve a work models which analyze the output of the base model to avoid inaccurate predictions of poorly trained models or redundant execution on multiple models or making remote application programming interface (API) calls to the remote system to perform an image classification task. See, Wen, paragraphs 12-13. However, none of the applied references appear to disclose "computing an approximation of a posterior probability using a variational distribution of weights based on the extracted set of features of the one or more personal examples." The other cited references also fail to disclose the identified portion of claim 1 and therefore, fail to remedy the deficiencies of Viswanathan, Huang and Wen.
Applicant’s arguments related to the cited limitation have been considered but are moot because the newly amended claim necessitated a new ground of rejection that does not rely on Viswanathan and Huang alone. Rather, Viswanathan and Huang in combination with newly applied prior art, Su, was used to address the said limitation.
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. Such claim limitations are:
Claim 17: “means for accessing… means for receiving… means for extracting… means for computing… means for generating”
Claim 18: “means for receiving… means for processing… means for generating…”
Claim 19: “means for computing…”
Claim 20: “means for calculating… means for sampling…”
Claim 21: “means for training…”
Claim 22: “means for receive… means for generate…”
Claim 23: “means for computing…”
3-Prong Analysis:
Prong (A):
In accordance with the MPEP, Prong (A) requires:
(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 ....
Examiner finds that claims 17-23 use the term “means”, therefore satisfying the Prong (A) of the 3-Prong analysis.
Prong (B):
In accordance with the MPEP, Prong (B) requires:
(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” ....
Based upon claims 17-23, the Examiner finds that the terms “accessing, receiving, extracting, computing, generating, processing, calculating, sampling, and training” are modified by functional language linked by transition word “for,” therefore satisfying the Prong (B) of the 3-Prong analysis.
Prong (C):
‘In accordance with the MPEP, Prong (C) requires:
(C) the term “means” or “step” or the generic placeholder is not modified by sufficient structure, material, or acts for performing the claimed function.
Based upon a review of claims 17-23, the Examiner finds that the term “means” is not modified by sufficient structure, material, or acts for performing the claimed function. Therefore, the Examiner concludes that cited claims meets invocation Prong (C) of the 3-Prong analysis.
Corresponding Structure and Algorithm
After a claimed phrase has been shown to invoke 35 U.S.C. § 112 (f), the next step is to determine the corresponding structure, material, or acts as described in the specification. MPEP § 2181. II. Furthermore, See MPEP 2161.01.
Computer/Component Support and Algorithm:
Upon reviewing Paragraph 0007, as well as drawings (Fig. 1) and associated description (Paragraphs 0029-0035), Examiner concludes that the cited limitations above disclose or are described in a way that one of ordinary skill in the art will understand what “structure” or “material” the inventor has identified to perform the recited functions.
Upon reviewing the disclosure and drawings, Examiner concludes that the “means for” limitations above perform the described functions of claims 17-23 using the algorithm and flowchart in Fig. 7, as well as associated paragraphs 0060-0069 and 0076-0080.
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 § 101
Claims 1-30 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Claims 1-8 are directed towards a computer-implemented method. Claims 9-23 are directed towards an apparatus. Claims 24-30 are directed towards a non-transitory computer readable medium. Therefore, all claims are directed towards one of the four statutory categories of patent eligible subject matter.
Claim 1
Step 2A Prong 1:
Claim 1 recites:
“extracting a set of features of the one or more personal examples;” Extracting a set of features of the one or more personal examples is an action that can be performed mentally with the aid of pen and paper, and is therefore a mental process.
“computing an approximation of a posterior probability [using a variational distribution of weights] based on the extracted set of features of the one or more personal examples;” Computing an approximation of a posterior probability based on the extracted set of features is a claim that merely uses textual replacements for particular equations, and is therefore a mathematical concept.
“generating, [at the edge device], a personal model [using the shared neural network model];” Generating a personal model is an action that can be performed mentally with the aid of pen and paper, and is therefore a mental process.
Step 2A Prong Two
This judicial exception is not integrated into a practical application because the additional elements are as follows:
“A computer-implemented method;” Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
“accessing a shared neural network model;” The limitation amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)).
“at an edge device;” Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
“receiving one or more personal examples of a first user at an edge device;” Mere data gathering recited at a high level of generality, and thus are insignificant extra-solution activity (MPEP 2106.05(g)).
“using a variational distribution of weights;” The limitation amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)).
“using the shared neural network model;” Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
“the personal model including a set of personalized weights determined at run-time based on the approximated posterior probability;” The limitation amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)).
Step 2B:
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because the additional elements are as follows:
“A computer-implemented method;” Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)) which cannot provide an inventive concept.
“accessing a shared neural network model;” The limitation amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)) and cannot provide an inventive concept.
“at an edge device;” Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)) which cannot provide an inventive concept.
“receiving one or more personal examples of a first user at an edge device;” Mere data gathering recited at a high level of generality, and thus are insignificant extra-solution activity. See MPEP 2106.05(g). The additional element of “receiving” does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of receiving steps amounts to no more than mere data gathering. This element amounts to receiving data over a network and are well-understood, routine, conventional activity. See MPEP 2106.05(d), subsection II (i). This cannot provide an inventive concept.
“using a variational distribution of weights;” The limitation amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)) and cannot provide an inventive concept.
“using the shared neural network model;” Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)) which cannot provide an inventive concept.
“the personal model including a set of personalized weights determined at run-time based on the approximated posterior probability;” The limitation amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)) and cannot provide an inventive concept.
Even when considered in combination, these additional elements represent mere instructions to apply an exception and therefore do not provide an inventive concept. The claim is ineligible.
Claim 2
Step 2A Prong Two
This judicial exception is not integrated into a practical application because the additional elements are as follows:
“receiving, by a personal model, one or more subsequent inputs;” Mere data gathering recited at a high level of generality, and thus are insignificant extra-solution activity (MPEP 2106.05(g)).
“processing the one or more subsequent inputs by the personal model including the set of personalized weights;” The limitation amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)).
“generating an inference based on the processing using the personal model;” The limitation amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)).
Step 2B:
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because the additional elements are as follows:
“receiving, by the personal model, one or more subsequent inputs;” Mere data gathering recited at a high level of generality, and thus are insignificant extra-solution activity. See MPEP 2106.05(g). The additional element of “receiving” does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of receiving steps amounts to no more than mere data gathering. This element amounts to receiving data over a network and are well-understood, routine, conventional activity. See MPEP 2106.05(d), subsection II (i). This cannot provide an inventive concept.
“processing the one or more subsequent inputs by the personal model including the set of personalized weights;” The limitation amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)) and cannot provide an inventive concept.
“generating an inference based on the processing using the personal model;” The limitation amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)) and cannot provide an inventive concept.
Even when considered in combination, these additional elements represent mere instructions to apply an exception and therefore do not provide an inventive concept. The claim is ineligible.
Claim 3
Step 2A Prong 1:
Claim 3 recites:
“the approximated posterior probability is computed based on a mean and variance relative to the one or more personal examples;” The approximated posterior probability is computed based on a mean and variance relative to the one or more personal examples is a claim that merely uses textual replacements for particular equations, and is therefore a mathematical concept.
Step 2A Prong Two and Step 2B:
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception under step 2B. Thus, the judicial exception is not integrated into a practical application (see MPEP 2106.04(d) I.), failing step 2A prong 2. The claim is ineligible.
Claim 4
Step 2A Prong 1:
Claim 4 recites:
“calculating a mean and a variance based on the extracted set of features;” Calculating a mean and a variance based on the extracted set of features is a claim that merely uses textual replacements for particular equations, and is therefore a mathematical concept.
“sampling weights of the shared neural network model based on the mean and the variance;” Sampling weights of the shared model based on the mean and the variance is an action that can be performed mentally with the aid of pen and paper, and is therefore a mental process.
Step 2A Prong Two and Step 2B:
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception under step 2B. Thus, the judicial exception is not integrated into a practical application (see MPEP 2106.04(d) I.), failing step 2A prong 2. The claim is ineligible.
Claim 5
Step 2A Prong Two
This judicial exception is not integrated into a practical application because the additional elements are as follows:
“the mean and the variance are computed during training;” The limitation amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)).
Step 2B:
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because the additional elements are as follows:
“in which the mean and the variance are computed during training;” The limitation amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)) and cannot provide an inventive concept.
Even when considered in combination, these additional elements represent mere instructions to apply an exception and therefore do not provide an inventive concept. The claim is ineligible.
Claim 6
Step 2A Prong Two
This judicial exception is not integrated into a practical application because the additional elements are as follows:
“the set of personalized weights are trained based on a model loss function includes a cross-entropy loss and a Kullback-Leibler divergence;” The limitation amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)).
Step 2B:
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because the additional elements are as follows:
“the personalized weights are trained based on a model loss function includes a cross-entropy loss and a Kullback-Leibler divergence;” The limitation amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)) and cannot provide an inventive concept.
Even when considered in combination, these additional elements represent mere instructions to apply an exception and therefore do not provide an inventive concept. The claim is ineligible.
Claim 7
Step 2A Prong 1:
Claim 7 recites:
“generating a second set of personalized weights corresponding to the second user;” Generating a second set of personalized weights corresponding to the second user is an action that can be performed mentally with the aid of pen and paper, and is therefore a mental process.
Step 2A Prong Two
This judicial exception is not integrated into a practical application because the additional elements are as follows:
“receiving one or more inputs corresponding to a second user;” Mere data gathering recited at a high level of generality, and thus are insignificant extra-solution activity (MPEP 2106.05(g)).
Step 2B:
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because the additional elements are as follows:
“receiving one or more inputs corresponding to a second user;” Mere data gathering recited at a high level of generality, and thus are insignificant extra-solution activity. See MPEP 2106.05(g). The additional element of “receiving” does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of receiving steps amounts to no more than mere data gathering. This element amounts to receiving data over a network and are well-understood, routine, conventional activity. See MPEP 2106.05(d), subsection II (i). This cannot provide an inventive concept.
Even when considered in combination, these additional elements represent mere instructions to apply an exception and therefore do not provide an inventive concept. The claim is ineligible.
Claim 8
Step 2A Prong 1:
Claim 8 recites:
“computing in a testing phase, an output based on the set of personalized weights;” Computing in a testing phase, an output based on the personalized weights is a claim that merely uses textual replacements for particular equations, and is therefore a mathematical concept.
Step 2A Prong Two and Step 2B:
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception under step 2B. Thus, the judicial exception is not integrated into a practical application (see MPEP 2106.04(d) I.), failing step 2A prong 2. The claim is ineligible.
Claim 9
Step 2A Prong 1:
Claim 9 recites:
“extract a set of features of the one or more personal examples;” Extracting a set of features of the one or more personal examples is an action that can be performed mentally with the aid of pen and paper, and is therefore a mental process.
“compute an approximation of a posterior probability [using a variational distribution of weights] based on the extracted set of features of the one or more personal examples;” Computing an approximation of a posterior probability based on the extracted set of features is a claim that merely uses textual replacements for particular equations, and is therefore a mathematical concept.
“generate, at an edge device, a personal model [using the shared neural network model];” Generating a personal model is an action that can be performed mentally with the aid of pen and paper, and is therefore a mental process.
Step 2A Prong Two
This judicial exception is not integrated into a practical application because the additional elements are as follows:
“An apparatus comprising: at least one memory; and at least one processor coupled to the at least one memory;” Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
“access a shared neural network model;” The limitation amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)).
“receive one or more personal examples of a first user at an edge device;” Mere data gathering recited at a high level of generality, and thus are insignificant extra-solution activity (MPEP 2106.05(g)).
“using a variational distribution of weights;” The limitation amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)).
“using the shared neural network model;” Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
“the personal model including a set of personalized weights determined at run-time based on the approximated posterior probability;” The limitation amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)).
Step 2B:
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because the additional elements are as follows:
“An apparatus comprising: at least one memory; and at least one processor coupled to the at least one memory;” Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)) which cannot provide an inventive concept.
“access a shared neural network model;” The limitation amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)) and cannot provide an inventive concept.
“receive one or more personal examples of a first user at an edge device;” Mere data gathering recited at a high level of generality, and thus are insignificant extra-solution activity. See MPEP 2106.05(g). The additional element of “receiving” does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of receiving steps amounts to no more than mere data gathering. This element amounts to receiving data over a network and are well-understood, routine, conventional activity. See MPEP 2106.05(d), subsection II (i). This cannot provide an inventive concept.
“using a variational distribution of weights;” The limitation amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)) and cannot provide an inventive concept.
“using the shared neural network model;” Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)) which cannot provide an inventive concept.
“the personal model including a set of personalized weights determined at run-time based on the approximated posterior probability;” The limitation amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)) and cannot provide an inventive concept.
Even when considered in combination, these additional elements represent mere instructions to apply an exception and therefore do not provide an inventive concept. The claim is ineligible.
Claims 10-16 are apparatus claims that recite similar limitations to claims 2-8. Therefore, claims 10-16 are rejected using the same rationale as claims 2-8.
Claim 17
Step 2A Prong 1:
Claim 17 recites:
“means for extracting a set of features of the one or more personal examples;” Extracting a set of features of the one or more inputs is an action that can be performed mentally with the aid of pen and paper, and is therefore a mental process.
“means for computing an approximation of a posterior probability [using a variational distribution of weights] based on the extracted set of features;” Computing an approximation of a posterior probability based on the extracted set of features is a claim that merely uses textual replacements for particular equations, and is therefore a mathematical concept.
“means for generating, [at the edge device], a personal model [using the shared neural network model];” Generating a personal model is an action that can be performed mentally with the aid of pen and paper, and is therefore a mental process.
Step 2A Prong Two
This judicial exception is not integrated into a practical application because the additional elements are as follows:
“An apparatus;” Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
“means for accessing a shared neural network model;” The limitation amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)).
“at the edge device;” Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
“means for receiving one or more personal examples of a first user at an edge device;” Mere data gathering recited at a high level of generality, and thus are insignificant extra-solution activity (MPEP 2106.05(g)).
“using a variational distribution of weights;” The limitation amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)).
“using the shared neural network model;” Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
“the personal model including a set of personalized weights determined at run-time based on the approximated posterior probability;” The limitation amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)).
Step 2B:
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because the additional elements are as follows:
“An apparatus;” Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)) which cannot provide an inventive concept.
“means for accessing a shared neural network model;” The limitation amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)) and cannot provide an inventive concept.
“at the edge device;” Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)) which cannot provide an inventive concept.
“means for receiving one or more personal examples of a first user at an edge device;” Mere data gathering recited at a high level of generality, and thus are insignificant extra-solution activity. See MPEP 2106.05(g). The additional element of “receiving” does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of receiving steps amounts to no more than mere data gathering. This element amounts to receiving data over a network and are well-understood, routine, conventional activity. See MPEP 2106.05(d), subsection II (i). This cannot provide an inventive concept.
“using a variational distribution of weights;” The limitation amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)) and cannot provide an inventive concept.
“using the shared neural network model;” Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)) which cannot provide an inventive concept.
“the personal model including a set of personalized weights determined at run-time based on the approximated posterior probability;” The limitation amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)) and cannot provide an inventive concept.
Even when considered in combination, these additional elements represent mere instructions to apply an exception and therefore do not provide an inventive concept. The claim is ineligible.
Claims 18-20 and 21-23 are apparatus claims that recite similar limitations to claims 2-4 and 6-8, respectively. Therefore, claims 18-20 and 21-23 are rejected using the same rationale as claims 2-4 and 6-8, respectively.
Claim 24
Step 2A Prong 1:
Claim 24 recites:
“[program code to] extract a set of features of the one or more personal examples;” Extracting a set of features of the one or more personal examples is an action that can be performed mentally with the aid of pen and paper, and is therefore a mental process.
“[program code to] compute an approximation of a posterior probability based on the extracted set of features;” Computing an approximation of a posterior probability based on the extracted set of features is a claim that merely uses textual replacements for particular equations, and is therefore a mathematical concept.
“[program code to] generate, [at the edge device], a personal model [using the shared neural network model];” Generating a personal model including a set of personalized weights based on the approximated posterior probability is an action that can be performed mentally with the aid of pen and paper, and is therefore a mental process.
Step 2A Prong Two
This judicial exception is not integrated into a practical application because the additional elements are as follows:
“A non-transitory computer readable medium having encoded thereon program code, the program code being executed by a processor;” Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
“program code to;” Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
“program code to access a shared neural network model;” The limitation amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)).
“at the edge device;” Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
“program code to receive one or more personal examples of a first user at an edge device;” Mere data gathering recited at a high level of generality, and thus are insignificant extra-solution activity (MPEP 2106.05(g)).
“using a variational distribution of weights;” The limitation amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)).
“using the shared neural network model;” Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
“the personal model including a set of personalized weights determined at run-time based on the approximated posterior probability;” The limitation amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)).
Step 2B:
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because the additional elements are as follows:
“A non-transitory computer readable medium having encoded thereon program code, the program code being executed by a processor;” Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)) which cannot provide an inventive concept.
“program code to;” Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)) which cannot provide an inventive concept.
“program code to access a shared neural network model;” The limitation amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)) and cannot provide an inventive concept.
“at the edge device;” Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)) which cannot provide an inventive concept.
“program code to receive one or more personal examples of a first user at an edge device;” Mere data gathering recited at a high level of generality, and thus are insignificant extra-solution activity. See MPEP 2106.05(g). The additional element of “receiving” does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of receiving steps amounts to no more than mere data gathering. This element amounts to receiving data over a network and are well-understood, routine, conventional activity. See MPEP 2106.05(d), subsection II (i). This cannot provide an inventive concept.
“using a variational distribution of weights;” The limitation amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)) and cannot provide an inventive concept.
“using the shared neural network model;” Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)) which cannot provide an inventive concept.
“the personal model including a set of personalized weights determined at run-time based on the approximated posterior probability;” The limitation amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)) and cannot provide an inventive concept.
Even when considered in combination, these additional elements represent mere instructions to apply an exception and therefore do not provide an inventive concept. The claim is ineligible.
Claims 25-27 and 28-30 are apparatus claims that recite similar limitations to claims 2-4 and 6-8, respectively. Therefore, claims 25-27 and 28-30 are rejected using the same rationale as claims 2-4 and 6-8, respectively.
Claim Rejections - 35 USC § 103
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.
Claim(s) 1-2, 9-10, 17-18, and 24-25 is/are rejected under 35 U.S.C. 103 as being unpatentable over Viswanathan et al. (US 20210073631 A1) in view of Huang (CN 112560849 A, see attached translation), Su et al. (COMPOUND VARIATIONAL AUTO-ENCODER, published 2019), hereinafter Su, and Ambardekar et al. (US 20180300603 A1), hereinafter Ambardekar.
Regarding claim 1, Viswanathan teaches,
A computer-implemented method comprising: accessing a neural network model [Para 0007, Neural network systems and related machine learning methods are provided that use a dual neural network architecture to determined epistemic and aleatoric uncertainties associated with predicted output data];
Receiving one or more personal examples of, a first user at an edge device [Para 0084, device 2500 may include one or more of computers, such as a laptop computer, a desktop computer; Para 0088, One or more input/output (I/O) device(s) 2512 may also communicate via a user interface (UI) controller 2514, which may connect with I/O device(s) 2512 either directly or through bus 2508; Para 0091, input/output device(s) 2512 can allow a user (such as a human annotator) to enter commands and information to device 2500… Examples of input device(s) 2512 include, for example, sensors, a keyboard, a cursor control device (e.g., a mouse), a microphone, a scanner, and any other input devices known in the art];
and generating, at the edge device, a set of personalized weights (Para 0079, The parameters (e.g., weights) of the ANN are adjusted) based on the approximated posterior probability (Para 0079, drawn from the posterior distribution) [Para 0079, The parameters (e.g., weights) of the ANN are adjusted to minimize this cost function… In the forward-pass, one or more samples of the vector {circumflex over (x)} of measurement data is drawn from the posterior distribution. It is used to evaluate the cost function. In the backward-pass, a gradient function of the weights of the ANN is calculated via backpropagation so that their values can be updated by optimization of the cost function].
Viswanathan does not teach wherein the neural network model is a shared neural network model, extracting a set of features of the one or more personal examples, computing an approximation of a posterior probability using a variational distribution of weights based on the extracted set of features, and a personal model using the shared neural network model, the personal model including a set of personalized weights determined at run-time.
Huang teaches,
wherein the neural network model is a shared neural network model [Para 0027, The shared sub-model may include a multi-layer convolutional neural network and a recurrent neural network];
extracting a set of features (Para 0045, first eigenvectors and second eigenvectors) of one or more personal examples (Para 0045, first intermediate image) [Para 0045, The first region of the first intermediate image and the second intermediate image are cut to obtain n row images, where n is a positive integer; first eigenvectors and second eigenvectors of the n row images are extracted];
computing an approximation of a posterior probability based on the extracted set of features (Para 0045, the first eigenvector and the second eigenvector) of one or more personal examples (Para 0045, first intermediate image) [Para 0045, the first eigenvector and the second eigenvector are spliced end to end to obtain a target splicing vector, and the target splicing vector is input into a shared sub-model of an image segmentation model to obtain a first posterior probability in each row image].
a personal model (Para 0045, image segmentation model) using the shared neural network model [Para 0045, the target splicing vector is input into a shared sub-model of an image segmentation model to obtain a first posterior probability in each row image, where the shared sub-model includes multiple neural network operations],
the personal model including a set of personalized weights [Para 0045, The calculation operation may specifically include:… dividing the weight matrix (dimension is m n) corresponding to the weight data into k column vectors… (if m is less than n, the weight matrix can be filled with all zeros first)].
Huang is analogous to the claimed invention as they both relate to neural network-based computations. Therefore, it would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Viswanathan’s teachings to incorporate the teachings of Huang and provide a shared neural network that extracts features and creates a posterior probability in order to obtain an updated likelihood that improves the generation of personalized weights.
Viswanathan-Huang do not teach computing an approximation using a variational distribution of weights; and weights determined at run-time.
Su teaches,
computing an approximation using a variational distribution of weights [Sect 2.1, To derive the learning objective of the proposed model, first, we define the joint generative distribution pθ(x, z, w) and the inference distribution qφ(z, w | x)… where z is the latent representation of VAE, w is the weights of BNN, and x is the input datapoint. In the variational inference approach, the goal is to find the parameters φ and θ that minimize KL divergence between the approximate distribution and the true posterior].
Su is analogous to the claimed invention as they both relate to Bayesian methodologies in Variational Auto-Encoders. Therefore, it would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Viswanathan and Huang’s teachings to incorporate the teachings of Su and provide computing an approximation using a variational distribution of weights in order to improve the robustness of uncertainty quantification.
Viswanathan-Huang-Su do not teach weights determined at run-time.
Ambardekar teaches,
weights determined at run-time [Para 0035, at an exemplary run time, inline vector dequantization processing can be performed to determine the underlying neuron weight values that can result in the maintenance of neuron throughput and maintain optimized performance of the NN].
Ambardekar is analogous to the claimed invention as they both relate to deep neural networks. Therefore, it would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Viswanathan, Huang, and Su’s teachings to incorporate the teachings of Ambardekar and provide weights determined at run-time in order to [Ambardekar, para 0035] maintain optimized performance of the NN.
Regarding claim 2, Viswanathan-Huang-Su-Ambardekar teach all the limitations of claim 1 including the personal model (Wen, para 12 and 32).
Viswanathan further teaches,
Receiving one or more subsequent (Para 0079, This process can be repeated) inputs [Para 0079, FIG. 8 illustrates a training iteration of the ANN where one or more samples of vector 2 of measurement data are supplied as training data (input) to the ANN… This process can be repeated until the minimization of the cost function converges];
processing the one or more subsequent inputs by the personal model (Para 0079, ANN) including the set of personalized weights (Para 0079, The parameters (e.g., weights) of the ANN are adjusted) [Para 0079, FIG. 8 illustrates a training iteration of the ANN where one or more samples of vector 2 of measurement data are supplied as training data (input) to the ANN. The predicted total uncertainty σ.sub.tot as output by the ANN and the error (or difference) between the ground-truth for the formation property data and the mean formation property data (ŷ) generated by the BNN for the same vector {circumflex over (x)} of measurement data are used to compute a cost function. The parameters (e.g., weights) of the ANN are adjusted to minimize this cost function. This process can be repeated until the minimization of the cost function converges];
and generating an inference (Para 0079, The predicted total uncertainty σ.sub.tot) based on the processing using the personal model [Para 0079, The predicted total uncertainty σ.sub.tot as output by the ANN].
Regarding claim 9, Viswanathan teaches,
An apparatus comprising: at least one memory; and at least one processor coupled to the at least one memory [Para 0082, FIG. 10 illustrates an example device 2500, with a processor 2502 and memory 2504 that can be configured to implement various embodiments of the neural network inference systems and associated training methods and workflows as discussed in this disclosure], the at least one processor being configured to:
Receive one or more personal examples of a first user at an edge device [Para 0091, input/output device(s) 2512 can allow a user (such as a human annotator) to enter commands and information to device 2500, and also allow information to be presented to the user and/or other components or devices…Examples of input device(s) 2512 include, for example, sensors, a keyboard, a cursor control device (e.g., a mouse), a microphone, a scanner, and any other input devices known in the art; Para 0084, device 2500 may include one or more of computers, such as a laptop computer, a desktop computer];
Access a neural network model; extract a set of features of the one or more personal examples and generate, at the edge device, based on the approximated posterior probability [Para 0079, The parameters (e.g., weights) of the ANN are adjusted to minimize this cost function… In the forward-pass, one or more samples of the vector {circumflex over (x)} of measurement data is drawn from the posterior distribution. It is used to evaluate the cost function. In the backward-pass, a gradient function of the weights of the ANN is calculated via backpropagation so that their values can be updated by optimization of the cost function].
Viswanathan does not teach wherein the neural network model is a shared neural network model, compute an approximation of a posterior probability using a variational distribution of weights based on the extracted set of features, and a personal model using the shared neural network model, the personal model including a set of personalized weights determined at run-time.
Huang teaches,
wherein the neural network model is a shared neural network model [Para 0027, The shared sub-model may include a multi-layer convolutional neural network and a recurrent neural network];
compute an approximation of a posterior probability based on the extracted set of features (Para 0045, the first eigenvector and the second eigenvector) of one or more personal examples (Para 0045, first intermediate image) [Para 0045, the first eigenvector and the second eigenvector are spliced end to end to obtain a target splicing vector, and the target splicing vector is input into a shared sub-model of an image segmentation model to obtain a first posterior probability in each row image]
a personal model (Para 0045, image segmentation model) using the shared neural network model [Para 0045, the target splicing vector is input into a shared sub-model of an image segmentation model to obtain a first posterior probability in each row image, where the shared sub-model includes multiple neural network operations],
the personal model including a set of personalized weights [Para 0045, The calculation operation may specifically include:… dividing the weight matrix (dimension is m n) corresponding to the weight data into k column vectors… (if m is less than n, the weight matrix can be filled with all zeros first)].
Huang is analogous to the claimed invention as they both relate to neural network-based computations. Therefore, it would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Viswanathan’s teachings to incorporate the teachings of Huang and provide a shared neural network that extracts features and creates a posterior probability in order to obtain an updated likelihood that improves the generation of personalized weights.
Viswanathan-Huang do not teach compute an approximation using a variational distribution of weights; and weights determined at run-time.
Su teaches,
computing an approximation using a variational distribution of weights [Sect 2.1, To derive the learning objective of the proposed model, first, we define the joint generative distribution pθ(x, z, w) and the inference distribution qφ(z, w | x)… where z is the latent representation of VAE, w is the weights of BNN, and x is the input datapoint. In the variational inference approach, the goal is to find the parameters φ and θ that minimize KL divergence between the approximate distribution and the true posterior].
Su is analogous to the claimed invention as they both relate to Bayesian methodologies in Variational Auto-Encoders. Therefore, it would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Viswanathan and Huang’s teachings to incorporate the teachings of Su and provide computing an approximation using a variational distribution of weights in order to improve the robustness of uncertainty quantification.
Viswanathan-Huang-Su do not teach weights determined at run-time.
Ambardekar teaches,
weights determined at run-time [Para 0035, at an exemplary run time, inline vector dequantization processing can be performed to determine the underlying neuron weight values that can result in the maintenance of neuron throughput and maintain optimized performance of the NN].
Ambardekar is analogous to the claimed invention as they both relate to deep neural networks. Therefore, it would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Viswanathan, Huang, and Su’s teachings to incorporate the teachings of Ambardekar and provide weights determined at run-time in order to [Ambardekar, para 0035] maintain optimized performance of the NN.
Claim 10 is an apparatus claim that recites similar limitations to claim 2. Therefore, claim 10 is rejected using the same rationale as claim 2.
Regarding claim 17 Viswanathan teaches,
An apparatus [Fig. 10] comprising: a means for accessing a neural network model [Para 0007, Neural network systems and related machine learning methods are provided that use a dual neural network architecture to determined epistemic and aleatoric uncertainties associated with predicted output data];
A means for receiving one or more personal examples of a first user at an edge device [Para 0091, input/output device(s) 2512 can allow a user (such as a human annotator) to enter commands and information to device 2500, and also allow information to be presented to the user and/or other components or devices…Examples of input device(s) 2512 include, for example, sensors, a keyboard, a cursor control device (e.g., a mouse), a microphone, a scanner, and any other input devices known in the art; Para 0084, device 2500 may include one or more of computers, such as a laptop computer, a desktop computer];
and a means for generating, at the edge device, based on the approximated posterior probability [Para 0079, The parameters (e.g., weights) of the ANN are adjusted to minimize this cost function… In the forward-pass, one or more samples of the vector {circumflex over (x)} of measurement data is drawn from the posterior distribution. It is used to evaluate the cost function. In the backward-pass, a gradient function of the weights of the ANN is calculated via backpropagation so that their values can be updated by optimization of the cost function; Para 0084, 0084, device 2500 may include one or more of computers, such as a laptop computer, a desktop computer].
Viswanathan does not teach wherein the neural network model is a shared neural network model, a means for extracting a set of features of the one or more personal examples, a means for computing an approximation of a posterior probability using a variational distribution of weights based on the extracted set of features, the personal model including a set of personalized weights determined at run-time.
Huang teaches,
wherein the neural network model is a shared neural network model [Para 0027, The shared sub-model may include a multi-layer convolutional neural network and a recurrent neural network.];
a means for extracting a set of features (Para 0045, first eigenvectors and second eigenvectors) of one or more personal examples (Para 0045, first intermediate image) [Para 0045, The first region of the first intermediate image and the second intermediate image are cut to obtain n row images, where n is a positive integer; first eigenvectors and second eigenvectors of the n row images are extracted];
a means for computing an approximation of a posterior probability based on the extracted set of features (Para 0045, the first eigenvector and the second eigenvector) of one or more personal examples (Para 0045, first intermediate image) [Para 0045, the first eigenvector and the second eigenvector are spliced end to end to obtain a target splicing vector, and the target splicing vector is input into a shared sub-model of an image segmentation model to obtain a first posterior probability in each row image]
a personal model (Para 0045, image segmentation model) using the shared neural network model [Para 0045, the target splicing vector is input into a shared sub-model of an image segmentation model to obtain a first posterior probability in each row image, where the shared sub-model includes multiple neural network operations],
the personal model including a set of personalized weights [Para 0045, The calculation operation may specifically include:… dividing the weight matrix (dimension is m n) corresponding to the weight data into k column vectors… (if m is less than n, the weight matrix can be filled with all zeros first)].
Huang is analogous to the claimed invention as they both relate to neural network-based computations. Therefore, it would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Viswanathan’s teachings to incorporate the teachings of Huang and provide a shared neural network that extracts features and creates a posterior probability in order to obtain an updated likelihood that improves the generation of personalized weights.
Viswanathan-Huang do not teach a means for computing an approximation using a variational distribution of weights; and weights determined at run-time.
Su teaches,
A means for computing an approximation using a variational distribution of weights [Sect 2.1, To derive the learning objective of the proposed model, first, we define the joint generative distribution pθ(x, z, w) and the inference distribution qφ(z, w | x)… where z is the latent representation of VAE, w is the weights of BNN, and x is the input datapoint. In the variational inference approach, the goal is to find the parameters φ and θ that minimize KL divergence between the approximate distribution and the true posterior].
Su is analogous to the claimed invention as they both relate to Bayesian methodologies in Variational Auto-Encoders. Therefore, it would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Viswanathan and Huang’s teachings to incorporate the teachings of Su and provide computing an approximation using a variational distribution of weights in order to improve the robustness of uncertainty quantification.
Viswanathan-Huang-Su do not teach weights determined at run-time.
Ambardekar teaches,
weights determined at run-time [Para 0035, at an exemplary run time, inline vector dequantization processing can be performed to determine the underlying neuron weight values that can result in the maintenance of neuron throughput and maintain optimized performance of the NN].
Ambardekar is analogous to the claimed invention as they both relate to deep neural networks. Therefore, it would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Viswanathan, Huang, and Su’s teachings to incorporate the teachings of Ambardekar and provide weights determined at run-time in order to [Ambardekar, para 0035] maintain optimized performance of the NN.
Claim 18 is an apparatus claim that recites similar limitations to claim 2. Therefore, claim 18 is rejected using the same rationale as claim 2.
Regarding claim 24 Viswanathan teaches,
A non-transitory computer readable medium having encoded thereon program code, the program code being executed by a processor [Para 0094, Such computer instructions can be stored in a non-transitory computer readable medium (e.g., memory) and executed by the computer processor] and comprising:
program code to receive one or more personal examples of a first user at an edge device [Para 0091, input/output device(s) 2512 can allow a user (such as a human annotator) to enter commands and information to device 2500, and also allow information to be presented to the user and/or other components or devices…Examples of input device(s) 2512 include, for example, sensors, a keyboard, a cursor control device (e.g., a mouse), a microphone, a scanner, and any other input devices known in the art; Para 0084, device 2500 may include one or more of computers, such as a laptop computer, a desktop computer];
program code to access a neural network model; program code to generate at the edge device, based on the approximated posterior probability [Para 0079, The parameters (e.g., weights) of the ANN are adjusted to minimize this cost function… In the forward-pass, one or more samples of the vector {circumflex over (x)} of measurement data is drawn from the posterior distribution. It is used to evaluate the cost function. In the backward-pass, a gradient function of the weights of the ANN is calculated via backpropagation so that their values can be updated by optimization of the cost function; Para 0084, device 2500 may include one or more of computers, such as a laptop computer, a desktop computer].
Viswanathan teaches the limitations of claim 24 including program code (Viswanathan, Para 0094).
Viswanathan does not teach wherein the neural network model is a shared neural network model, to extract a set of features of the one or more personal examples, to compute an approximation of a posterior probability using a variational distribution of weights based on the extracted set of features, and a personal model using the shared neural network model, the personal model including a set of personalized weights determined at run-time.
Huang teaches,
wherein the neural network model is a shared neural network model [Para 0027, The shared sub-model may include a multi-layer convolutional neural network and a recurrent neural network];
to extract a set of features (Para 0045, first eigenvectors and second eigenvectors) of one or more personal examples (Para 0045, first intermediate image) [Para 0045, The first region of the first intermediate image and the second intermediate image are cut to obtain n row images, where n is a positive integer; first eigenvectors and second eigenvectors of the n row images are extracted];
to compute an approximation of a posterior probability (Para 0045, first posterior probability) based on the extracted set of features (Para 0045, the first eigenvector and the second eigenvector) of one or more personal examples (Para 0045, first intermediate image) [Para 0045, the first eigenvector and the second eigenvector are spliced end to end to obtain a target splicing vector, and the target splicing vector is input into a shared sub-model of an image segmentation model to obtain a first posterior probability in each row image]
a personal model (Para 0045, image segmentation model) using the shared neural network model [Para 0045, the target splicing vector is input into a shared sub-model of an image segmentation model to obtain a first posterior probability in each row image, where the shared sub-model includes multiple neural network operations],
the personal model including a set of personalized weights [Para 0045, The calculation operation may specifically include:… dividing the weight matrix (dimension is m n) corresponding to the weight data into k column vectors… (if m is less than n, the weight matrix can be filled with all zeros first)].
Huang is analogous to the claimed invention as they both relate to neural network-based computations. Therefore, it would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Viswanathan’s teachings to incorporate the teachings of Huang and provide a shared neural network that extracts features and creates a posterior probability in order to obtain an updated likelihood that improves the generation of personalized weights.
Viswanathan-Huang do not teach to compute an approximation using a variational distribution of weights; and weights determined at run-time.
Su teaches,
To compute an approximation using a variational distribution of weights [Sect 2.1, To derive the learning objective of the proposed model, first, we define the joint generative distribution pθ(x, z, w) and the inference distribution qφ(z, w | x)… where z is the latent representation of VAE, w is the weights of BNN, and x is the input datapoint. In the variational inference approach, the goal is to find the parameters φ and θ that minimize KL divergence between the approximate distribution and the true posterior].
Su is analogous to the claimed invention as they both relate to Bayesian methodologies in Variational Auto-Encoders. Therefore, it would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Viswanathan and Huang’s teachings to incorporate the teachings of Su and provide computing an approximation using a variational distribution of weights in order to improve the robustness of uncertainty quantification.
Viswanathan-Huang-Su do not teach weights determined at run-time.
Ambardekar teaches,
weights determined at run-time [Para 0035, at an exemplary run time, inline vector dequantization processing can be performed to determine the underlying neuron weight values that can result in the maintenance of neuron throughput and maintain optimized performance of the NN].
Ambardekar is analogous to the claimed invention as they both relate to deep neural networks. Therefore, it would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Viswanathan, Huang, and Su’s teachings to incorporate the teachings of Ambardekar and provide weights determined at run-time in order to [Ambardekar, para 0035] maintain optimized performance of the NN.
Claim 25 is a non-transitory computer readable medium claim that recites similar limitations to claim 2. Therefore, claim 25 is rejected using the same rationale as claim 2.
Claim(s) 3, 11, 19, and 26 is/are rejected under 35 U.S.C. 103 as being unpatentable over Viswanathan in view of Huang, Su, and Ambardekar, and in further view of Yan (CN 112634171 A, see attached translation).
Regarding claim 3, Viswanathan-Huang-Su-Ambardekar teach all the limitations of claim 1.
Viswanathan-Huang-Su-Ambardekar do not teach the approximated posterior probability is computed based on a mean and variance relative to the one or more personal examples.
Yan teaches,
approximated posterior probability is computed based on a mean and variance relative to one or more personal examples [Para 0009, Step S2: input the training set into a Bayesian convolutional neural network, and train the Bayesian convolutional neural network to obtain an optimal model; Para n0018, Use the probability distribution q(ω|θ) to approximate the posterior probability p(W|D); Para n0019, Where θ＝(μ,σ), θ represents a set with mean μ and standard deviation σ, and ω represents weight].
Yan is analogous to the claimed invention as they both relate to applying Bayesian methodologies to neural networks. Therefore, it would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Viswanathan, Huang, Su, and Ambardekar’s teachings to incorporate the teachings of Yan and provide the approximated posterior probability being computed based on a mean and variance to make the probability more accurate, resulting in enhanced weight updates and improved predictions.
Claims 11, 19, and 26 are apparatus and non-transitory computer readable medium claims that recite similar limitations to claim 3. Therefore, claims 11, 19, and 26 are rejected using the same rationale as claim 3.
Claim(s) 4-5, 12-13, 20, and 27 is/are rejected under 35 U.S.C. 103 as being unpatentable over Viswanathan in view of Huang, Su, and Ambardekar, and in further view of Zhai et al. (CN 108009472 A, see attached translation) and Yan.
Regarding claim 4, Viswanathan-Huang-Su-Ambardekar teach all the limitations of claim 1 including the shared neural network model (claim 1: Huang, Para 0027).
Viswanathan-Huang-Su-Ambardekar do not teach calculating a mean and a variance based on the extracted set of features; and sampling weights of the shared model based on the mean and the variance.
Zhai teaches,
calculating a mean and a variance based on the extracted set of features [Para 0074, S6), input the image feature vector A = (a<sub>1</sub>, a<sub>2</sub>, .... a<sub>m</sub>) extracted in step S5) and the category set C = (y<sub>1</sub>, y<sub>2</sub>, .... y<sub>n</sub>) into the Bayesian classifier model for training, and count the conditional probability P(a<sub>i</sub>|y<sub>j</sub>) of the feature attribute a<sub>i</sub> under each category y<sub>j</sub>. Since the feature attribute is a continuous value, it is assumed that its value obeys a Gaussian distribution… thereby obtaining the mean and standard deviation of the corresponding features under this category];
Zhai is analogous to the claimed invention as they both relate to applying Bayesian techniques to neural networks. Therefore, it would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Viswanathan, Huang, Su, and Ambardekar’s teachings to incorporate the teachings of Zhai and provide calculating mean and variance from extracted feature to improve the calculation of the means for updating weights.
Viswanathan-Huang-Su-Ambardekar-Zhai do not teach sampling weights of the shared model based on the mean and the variance.
Yan further teaches,
sampling weights (Para n0019, the weight ω<sub>i</sub> of the i-th group of data is sampled) based on the mean (Para n0019, μ<sub>i</sub>) and the variance (Para n0019, σ<sub>i</sub>) [Para n00018-n0019, Use the probability distribution q(ω|θ) to approximate the posterior probability p(W|D); Where θ＝(μ,σ), θ represents a set with mean μ and standard deviation σ, and ω represents weight; the weight ω<sub>i</sub> of the i-th group of data is sampled from the normal distribution (μ<sub>i</sub>,σ<sub>i</sub>), i represents the i-th group of data, μ<sub>i</sub> represents the mean of the i-th group of data, and σ<sub>i</sub> represents the standard deviation of the i-th group of data].
Yan is analogous to the claimed invention as they both relate to applying Bayesian techniques to neural networks. Therefore, it would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Viswanathan, Huang, Wen, and Yan’s teachings to incorporate the teachings of Yan and provide calculating mean and variance from extracted feature to improve the calculation of the means for updating weights.
Regarding claim 5, Viswanathan-Huang-Su-Ambardekar-Zhai-Yan teach all the limitations of claim 4.
Viswanathan further teaches,
Mean (Para 0019, a mean value output) and the variance (Para 0019, epistemic uncertainty of the output) are computed during training [Para 0019, a machine learning method is provided that involves a training phase and an inference phase. In the training phase, a first neural network is trained to predict a mean value output and epistemic uncertainty of the output given input data, and a second neural network is trained to predict total uncertainty of the output of the first neural network; Para 0039, The BNN is configured to predict or estimate a mean output and associated standard deviation (epistemic uncertainty)].
Claims 12, 20, and 27 are apparatus and non-transitory computer readable medium claims that recite similar limitations to claim 4. Therefore, claims 12, 20, and 27 are rejected using the same rationale as claim 4.
Claims 13 is an apparatus claim that recites similar limitations to claim 5. Therefore, claim 13 is rejected using the same rationale as claim 5.
Claim(s) 6, 14, 21, and 28 is/are rejected under 35 U.S.C. 103 as being unpatentable over Viswanathan in view of Huang, Su, and Ambardekar, and in further view of Oroojlooyjadid et al. (US 20220374732 A1), hereinafter Oroojlooyjadid, and Tanimoto et al. (US 20200005183 A1), hereinafter Tanimoto.
Regarding claim 6, Viswanathan-Huang-Su-Ambardekar teach all the limitations of claim 1 including the personalized weights (claim 1: Visawanathan, Para 0079).
Oroojlooyjadid teaches,
in which set of weights are trained based on a model loss function includes a cross-entropy loss [Para 0082, The training parameters for the AE model may be associated with application of gradient descent in updating gradient vectors and weight vectors each iteration, with application of forward and backward propagation, with a loss function, etc. For example, a binary cross-entropy (BCE) loss function may be used to train the AE model]
Oroojlooyjadid is analogous to the claimed invention as they both relate to training neural network weights. Therefore, it would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Viswanathan, Huang, Su, and Ambardekar’s teachings to incorporate the teachings of Oroojlooyjadid and provide weights trained on a cross-entropy loss in order to attain weights that generate better prediction outcomes.
Viswanathan-Huang-Su-Ambardekar-Oroojlooyjadid do not teach a model loss function includes a Kullback-Leibler divergence.
Tanimoto teaches,
a model loss function includes a Kullback-Leibler divergence [Para 0106, A loss function formed of… KL-Divergence (Kullback-Leibler-Divergence)].
Tanimoto is analogous to the claimed invention as they both relate to training neural network weights. Therefore, it would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Viswanathan, Huang, Wen, and Oroojlooyjadid’s teachings to incorporate the teachings of Tanimoto and provide Kullback-Leibler divergence in order to attain weights that generate better prediction outcomes.
Claims 14, 21, and 28 are apparatus and non-transitory computer readable medium claims that recite similar limitations to claim 6. Therefore, claims 14, 21, and 28 are rejected using the same rationale as claim 6.
Claim(s) 7-8, 15-16, 22-23, 29-30 and is/are rejected under 35 U.S.C. 103 as being unpatentable over Viswanathan in view of Huang and Wen, and in further view of Chen (US 20150154508 A1), hereinafter Chen.
Regarding claim 7, Viswanathan-Huang-Su-Ambardekar teach all the limitations of claim 1.
Viswanathan-Huang-Su-Ambardekar do not teach receiving one or more inputs corresponding to a second user; and generating a second set of personalized weights corresponding to the second user.
Chen teaches,
receiving one or more inputs corresponding to a second user (Para 0027, input by the user); and generating (Para 0014, trained) a second set of personalized weights (Para 0014, individualized weight) corresponding to the second user (Para 0014, each user) [Para 0014, The individualized weight of the characteristic of each data object with respect to the characteristic of each user is trained according to the satisfaction degree of each user behavior data, the characteristic of the data object, and the characteristic of the user recorded in each user behavior data; Para 0027, When conducting a data search according to a query word input by the user, with respect to found one or more data objects, the present techniques, according to the individualized weight of each characteristic combination, find a corresponding individualized weight of the characteristics of the user and the characteristic of each data object and calculate an individualized score of each data object searched by the user].
Chen is analogous to the claimed invention as they both relate to predictions via machine learning models. Therefore, it would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Viswanathan, Huang, Su, and Ambardekar’s teachings to incorporate the teachings of Chen and provide a second user that generates a second set of weights in order to improve a model’s prediction accuracy by producing weights tailored to data observed by different users.
Regarding claim 8, Viswanathan-Huang-Su-Ambardekar teach all the limitations of claim 1.
Viswanathan further teaches,
Computing in a testing phase (Para 0064, inference phase), an output (Para 0064, the mean value and the epistemic uncertainty) based on the set of weights (Para 0078, weights of the BNN) [Para 0064, In the inference phase, the same input data can be applied to both the BNN and the ANN simultaneously or in a parallel manner. The BNN estimates the mean value and the epistemic uncertainty σ.sub.1 of the unknown output ŷ; Para 0078, In the backward-pass, gradient functions of the probability distribution of weights of the BNN are calculated via backpropagation so that their values can be updated by optimization of the cost function.].
Viswanathan-Huang-Su-Ambardekar do not teach the weights being personalized weights.
Chen further teaches,
the weights being personalized weights [Para 0078, the individualized model is trained to obtain the individualized weight of the each characteristic or characteristic combination]
Chen is analogous to the claimed invention as they both relate to predictions via machine learning models. Therefore, it would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Viswanathan, Huang, Su, and Ambardekar’s teachings to incorporate the teachings of Chen and provide a personalized weights improve a model’s prediction accuracy by producing weights tailored to data observed by different users.
Claims 15-16, 22-23, and 29-30 are apparatus and non-transitory computer readable medium claims that recite similar limitations to claims 7-8, respectively. Therefore, claims 15-16, 22-23, and 29-30 are rejected using the same rationale as claim 7-8, respectively.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to SYED RAYHAN AHMED whose telephone number is (571)270-0286. The examiner can normally be reached Mon-Fri ET.
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, David Yi can be reached at (571) 270-7519. 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.
/SYED RAYHAN AHMED/Examiner, Art Unit 2126
/VAN C MANG/Primary Examiner, Art Unit 2126