Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
This action is in response to the amendment and remarks filed 3/13/2026. In the amendment, claims 1-3, 5-6, 8-10, 12-13, 15-16, and 18-19 were amended and no claims were added. As such, claims 1-20 are pending.
The following new grounds of rejection are necessitated by the amendment submitted 3/13/2026.
Information Disclosure Statement
The information disclosure statements (IDS) submitted on April 9th, 2026 and February 5th, 2026 are in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statements are considered by the examiner.
Response to Arguments
Applicant’s arguments, see Page 13, filed March 13th, 2026, with respect to the double patenting rejections, the claim objections, and specification objections have been fully considered and are persuasive. The specification objection has been withdrawn. The objected claims and double patenting rejections of claims 1-20 has been withdrawn.
Applicant’s arguments with respect to the rejections of claims 1-20 under 35 U.S.C § 101 and 103 are not persuasive for the following reasons:
35 U.S.C. 101:
Applicant argues that the amended claims are directed to an improvement in machine learning technology, specifically a bias-reduced support vector machine (SVM) using a soft margin and a Kullback-Leibler (KL) divergence-based debiasing term, and therefore integrates any alleged judicial exception into a practical application under Step 2A Prong Two (See Pages 9-12 of Remarks). Applicant further relies on Ex Parte Desjardins (PTAB September 26th, 2025), Enfish (Federal Circuit 2016), and McRO (Federal Circuit 2016) to assert that the claims improve model trustworthiness and machine learning functionality. The Examiner respectfully disagrees.
As set forth in the 35 U.S.C. 101 rejections below, the core of the claimed invention is the training of a SVM using a soft-margin optimization objective modified with a KL-divergence-based term. This is a mathematical optimization technique that adjusts model parameters and is characterized as a mathematical concept. The SVM applies a soft margin and a debiasing term that uses a KL divergence to train the machine learning model, as seen in Figure 2C. The alleged “reduction in latent bias” is a result of applying this KL divergence and does not amount to a specific technological improvement to the functioning of the computer or to the SVM architecture itself.
With respect to Applicant’s reliance on Desjardins, Enfish, and McRO, the Examiner notes that unlike those cases, the present claims do not recite a specific improvement to computer functionality or a particular technological implementation of machine learning models.
First, unlike Applicant’s claims, the claims at issue in Desjardins, Enfish, and McRO are directed to specific improvements in computer functionality or technological processes, rather than to the mere application of mathematical concepts to perform generic machine learning tasks.
In Enfish, the claims were directed to a specific self-referential table that improved the way a computer stores and retrieves data, thereby improving computer functionality itself (See Enfish, LLC v. Microsoft Corp.). In McRO, the claims recited specific rules that automated a previously manual animation process in a manner that improved the technological process itself, rather than merely using a computer as a tool (See McRO, Inc. v. Bandai Namco Games America Inc.). Similarly, Desjardins involved a specific technological improvement in how a machine learning system was structured and operated to achieve improved functionality.
By contrast, the instant claims are directed to training a support vector machine using a soft-margin optimization modified by a Kullback-Leibler (KL) divergence-based debiasing term, which constitutes a mathematical concept. The alleged improvement of reducing the latent bias is achieved solely through the application of this mathematical formulation and does not reflect any improvement to the functioning of the computer itself, not to the SVM architecture.
Further, unlike the claims in Enfish and McRO, the present claims do not recite any specific data structure, rule-based mechanism, or technological implementation that changes how the computer operates. Instead, the claims merely apply a known mathematical technique within a generic machine learning pipeline.
Additionally, the remaining limitations of “identifying a condition,” “obtaining the model,” and “providing services using the inference,” merely amount to data gathering and post-solution activity performed on a generic computer, which does not integrate the judicial exception into a practical application (See MPEP 2106.05(g)).
Therefore, the claims do not integrate the judicial exception into a practical application under Step 2A Prong 2, and the rejection under 35 U.S.C. 101 is maintained.
35 U.S.C. 103:
Applicant argues that none of the references teach the amended limitations of using a debiasing term that utilizes a KL divergence (See Pages 12 and 13 of Remarks). Applicant’s arguments with respect to claim(s) 1-20 have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Claim 1
Step 1: The claim recites a method; therefore, it is directed to the statutory category of processes.
Step2A Prong 1: The claim recites, inter alia:
[I]dentifying an occurrence of a condition that indicates an inference is necessary to… based on the occurrence: This limitation is a mental process because it deals with the evaluation/judgement/opinion of identifying when a condition is met.
training a support vector machine-based inference model based on a soft margin and a debiasing term to obtain a latent bias reduced support vector machine-based inference model, wherein a first amount of latent bias exhibited in first inferences generated by the latent bias reduced support vector machine-based inference model is lower than a second amount of latent bias exhibited in second inferences generated by the support vector machine-based inference model, and the debiasing term utilizes a Kullback-Leibler (KL) divergence: This limitation recites a mathematical concept because it describes training a support vector machine by directly applying the soft-margin optimization and KL-divergence-based debias term as showing in Figure 2C, which is a mathematical objective function defining how the model parameters are computed and adjusted.
Step 2A Prong 2: This judicial exception is not integrated into a practical application because the additional elements are as follows:
[P]roviding computer implemented services using inference models… provide the computer implemented services… and providing the computer implemented services using the inference: 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)).
obtaining the latent bias reduced support vector machine-based inference model: Mere data gathering recited at a high level of generality, and thus is an insignificant extra-solution activity (MPEP 2106.05(g)).
obtaining the inference using the latent bias reduced support vector machine-based inference model, the inference being one of the first inferences generated by the latent bias reduced support vector machine- based inference model: Mere data gathering recited at a high level of generality, and thus is an 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:
[P]roviding computer implemented services using inference models… provide the computer implemented services… and providing the computer implemented services using the inference: 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 and cannot provide inventive concept (MPEP 2106.05(f)).
obtaining the latent bias reduced support vector machine-based inference model: 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.
obtaining the inference using the latent bias reduced support vector machine-based inference model, the inference being one of the first inferences generated by the latent bias reduced support vector machine- based inference model: The additional element of “obtaining” 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.
The elements in combination as an ordered whole still do not amount to significantly more than the judicial exception. The claim merely applies the abstract idea using generic computer implementation to perform SVM-based inference and provide computer-implemented services based on the results. The additional elements, including obtaining the model and generating inferences, are recited at a high level of generality and reflect routine machine learning operations without improvement to computer functionality or the SVM. Considered as a whole, the claim merely implements a known mathematical technique using conventional computing resources to achieve a desired output, which does not provide an inventive concept.
Therefore, the claim as a whole remains focused on the abstract idea and fails Step 2B of the eligibility analysis.
Claim 2
Step 1: A process, as above.
Step2A Prong 1: This claim does not recite any abstract ideas, but depends on base claim 1, which does.
Step 2A Prong 2: This judicial exception is not integrated into a practical application because the additional elements are as follows:
reading the latent bias reduced support vector machine-based inference model from storage: 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:
reading the latent bias reduced support vector machine-based inference model from storage: The additional element of “reading” 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, this additional element represents mere instructions to apply an exception and therefore does not provide an inventive concept. The claim is ineligible.
Claim 3
Step 1: A process, as above.
Step2A Prong 1: The claim recites, inter alia:
Step 2A Prong 2: This judicial exception is not integrated into a practical application because the additional elements are as follows:
the support vector machine-based inference model is further trained using training data: 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)).
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 support vector machine-based inference model is further trained using training data: 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 and cannot provide inventive concept (MPEP 2106.05(f)).
Even when considered in combination, this additional element represents mere instructions to apply an exception and therefore does not provide an inventive concept. The claim is ineligible.
Claim 4
Step 1: A process, as above.
Step2A Prong 1: This claim does not recite an additional abstract idea, but the claim depends on claim 3 which recites an abstract idea.
Step 2A Prong 2: This judicial exception is not integrated into a practical application because the additional elements are as follows:
the training data comprises: records, and each of the records comprises: at least one feature value; at least one label value associated with the at least one feature value; and at least one bias feature value associated with the at least one feature value: The limitation merely describes the type of data being processed and thus 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 training data comprises: records, and each of the records comprises: at least one feature value; at least one label value associated with the at least one feature value; and at least one bias feature value associated with the at least one feature value: The limitation merely describes the type of data being processed and thus 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 which cannot provide inventive concept (MPEP 2106.05(h)).
Even when considered in combination, this additional element represents mere instructions to apply an exception and therefore does not provide an inventive concept. The claim is ineligible.
Claim 5
Step 1 A process, as above.
Step2A Prong 1: The claim recites inter alia:
training the support vector machine-based inference model to obtain the latent bias reduced support vector machine-based inference model comprises: obtaining, based on the training data and an objective function based in part on the debiasing term and the soft margin, a decision boundary: This limitation recites a mathematical concept because it recites training a SVM model, which is a statistical learning algorithm that optimizes a mathematical objective function using training data.
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.
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 1: A process, as above.
Step2A Prong 1: The claim recites inter alia:
the debiasing term provides an incentive for a uniform distribution of the records with respect to one or more bias features across the decision boundary in the objective function: This limitation encompasses a mathematical concept because the SVM is trained using the debiasing term which uses a uniform distribution, which incorporates a statistical distribution in the training 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.
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 1: A process, as above.
Step2A Prong 1: The claim recites inter alia:
the objective function comprises a weight that scales a level of the incentive for the uniform distribution of the records: This limitation encompasses a mathematical concept because the SVM is trained using the objective function which uses a uniform distribution, which is a statistical distribution used in the training 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.
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 1: The claim recites a non-transitory machine-readable medium; therefore, it is directed to the statutory category of article of manufacture.
Step2A Prong 1: The claim recites, inter alia:
[I]dentifying an occurrence of a condition that indicates an inference is necessary to… based on the occurrence: This limitation is a mental process because it deals with the evaluation/judgement/opinion of identifying when a condition is met.
training a support vector machine-based inference model based on a soft margin and a debiasing term to obtain a latent bias reduced support vector machine-based inference model, wherein a first amount of latent bias exhibited in first inferences generated by the latent bias reduced support vector machine-based inference model is lower than a second amount of latent bias exhibited in second inferences generated by the support vector machine-based inference model, and the debiasing term utilizes a Kullback-Leibler (KL) divergence: This limitation recites a mathematical concept because it describes training a support vector machine by directly applying the soft-margin optimization and KL-divergence-based debias term as showing in Figure 2C, which is a mathematical objective function defining how the model parameters are computed and adjusted.
Step 2A Prong 2: This judicial exception is not integrated into a practical application because the additional elements are as follows:
A non-transitory machine-readable medium having instructions stored therein, which when executed by a processor, cause the processor to perform operations: 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)).
[P]roviding computer implemented services using inference models… provide the computer implemented services… and providing the computer implemented services using the inference: 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 and cannot provide inventive concept (MPEP 2106.05(f)).
obtaining the latent bias reduced support vector machine-based inference model: Mere data gathering recited at a high level of generality, and thus is an insignificant extra-solution activity (MPEP 2106.05(g)).
obtaining the inference using the latent bias reduced support vector machine-based inference model, the inference being one of the first inferences generated by the latent bias reduced support vector machine- based inference model: Mere data gathering recited at a high level of generality, and thus is an 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:
A non-transitory machine-readable medium having instructions stored therein, which when executed by a processor, cause the processor to perform operations: 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 and cannot provide inventive concept (MPEP 2106.05(f)).
[P]roviding computer implemented services using inference models… provide the computer implemented services… and providing the computer implemented services using the inference: 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 and cannot provide inventive concept (MPEP 2106.05(f)).
obtaining the latent bias reduced support vector machine-based inference model: 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.
obtaining the inference using the latent bias reduced support vector machine-based inference model, the inference being one of the first inferences generated by the latent bias reduced support vector machine- based inference model: The additional element of “obtaining” 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.
The elements in combination as an ordered whole still do not amount to significantly more than the judicial exception. The claim merely applies the abstract idea using generic computer implementation to perform SVM-based inference and provide computer-implemented services based on the results. The additional elements, including obtaining the model and generating inferences, are recited at a high level of generality and reflect routine machine learning operations without improvement to computer functionality or the SVM. Considered as a whole, the claim merely implements a known mathematical technique using conventional computing resources to achieve a desired output, which does not provide an inventive concept.
Therefore, the claim as a whole remains focused on the abstract idea and fails Step 2B of the eligibility analysis.
Claim 9 is an article of manufacture claim that recites similar limitations to method claim 2. Therefore, claim 9 is rejected using the same rationale as claim 2.
Claim 10 is an article of manufacture claim that recites similar limitations to method claim 3. Therefore, claim 10 is rejected using the same rationale as claim 3.
Claim 11 is an article of manufacture claim that recites similar limitations to method claim 4. Therefore, claim 11 is rejected using the same rationale as claim 4.
Claim 12 is an article of manufacture that recites similar limitations to method claim 5. Therefore, claim 12 is rejected using the same rationale as claim 5.
Claim 13 is an article of manufacture that recites similar limitations to method claim 6. Therefore, claim 13 is rejected using the same rationale as claim 6.
Claim 14 is an article of manufacture claim that recites similar limitations to method claim 7. Therefore, claim 14 is rejected using the same rationale as claim 7.
Claim 15
Step 1: The claim recites a data processing system; therefore, it is directed to the statutory category of machine.
Step2A Prong 1: The claim recites, inter alia:
[I]dentifying an occurrence of a condition that indicates an inference is necessary to… based on the occurrence: This limitation is a mental process because it deals with the evaluation/judgement/opinion of identifying when a condition is met.
training a support vector machine-based inference model based on a soft margin and a debiasing term to obtain a latent bias reduced support vector machine-based inference model, wherein a first amount of latent bias exhibited in first inferences generated by the latent bias reduced support vector machine-based inference model is lower than a second amount of latent bias exhibited in second inferences generated by the support vector machine-based inference model, and the debiasing term utilizes a Kullback-Leibler (KL) divergence: This limitation recites a mathematical concept because it describes training a support vector machine by directly applying the soft-margin optimization and KL-divergence-based debias term as showing in Figure 2C, which is a mathematical objective function defining how the model parameters are computed and adjusted.
Step 2A Prong 2: This judicial exception is not integrated into a practical application because the additional elements are as follows:
A data processing system, comprising: a processor; and a memory coupled to the processor to store instructions, which when executed by the processor, cause the processor to perform operations: 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)).
[P]roviding computer implemented services using inference models… provide the computer implemented services… and providing the computer implemented services using the inference: 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)).
obtaining the latent bias reduced support vector machine-based inference model: Mere data gathering recited at a high level of generality, and thus is an insignificant extra-solution activity (MPEP 2106.05(g)).
obtaining the inference using the latent bias reduced support vector machine-based inference model, the inference being one of the first inferences generated by the latent bias reduced support vector machine- based inference model: Mere data gathering recited at a high level of generality, and thus is an 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:
A data processing system, comprising: a processor; and a memory coupled to the processor to store instructions, which when executed by the processor, cause the processor to perform operations: 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 and cannot provide inventive concept (MPEP 2106.05(f)).
[P]roviding computer implemented services using inference models… provide the computer implemented services… and providing the computer implemented services using the inference: 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 and cannot provide inventive concept (MPEP 2106.05(f)).
obtaining the latent bias reduced support vector machine-based inference model: 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.
obtaining the inference using the latent bias reduced support vector machine-based inference model, the inference being one of the first inferences generated by the latent bias reduced support vector machine- based inference model: The additional element of “obtaining” 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.
The elements in combination as an ordered whole still do not amount to significantly more than the judicial exception. The claim merely applies the abstract idea using generic computer implementation to perform SVM-based inference and provide computer-implemented services based on the results. The additional elements, including obtaining the model and generating inferences, are recited at a high level of generality and reflect routine machine learning operations without improvement to computer functionality or the SVM. Considered as a whole, the claim merely implements a known mathematical technique using conventional computing resources to achieve a desired output, which does not provide an inventive concept.
Therefore, the claim as a whole remains focused on the abstract idea and fails Step 2B of the eligibility analysis.
Claim 16 is a machine claim that recites similar limitations to method claim 3. Therefore, claim 16 is rejected using the same rationale as claim 3.
Claim 17 is a machine claim that recites similar limitations to method claim 4. Therefore, claim 17 is rejected using the same rationale as claim 4.
Claim 18 is a machine claim that recites similar limitations to method claim 5. Therefore, claim 18 is rejected using the same rationale as claim 5.
Claim 19 is a machine claim that recites similar limitations to method claim 6. Therefore, claim 19 is rejected using the same rationale as claim 6.
Claim 20 is a machine claim that recites similar limitations to method claim 7. Therefore, claim 20 is rejected using the same rationale as claim 7.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 1-20 are rejected under 35 U.S.C. 103 as being unpatentable over Zafar (“Fairness Constraints: A Flexible Approach for Fair Classification”, 2017) in view of Gauci (“US 10831339 B2”) in view of Wu (“Incorporating Prior Knowledge with Weighted Margin Support Vector Machines”, 2004), and in further view of Buyl (“The KL-Divergence between a Graph Model and its Fair I-Projection as a Fairness Regularizer”, 2021).
Regarding claim 1,
Zafar teaches training a support vector machine-based inference model based on… a debiasing term to obtain a latent bias reduced support vector machine-based inference model (Page 1 Introduction of Zafar, “In this work, our goal is to design classifiers— specifically, convex margin-based classifiers like logistic regression and support vector machines (SVMs)—that avoid both disparate treatment and disparate impact…”, Page 4 Section 3.1, “Appendix A presents the specialization of our formulation for both linear and non-linear SVM classifiers.”, Page 2 Introduction, “we introduce a novel intuitive measure of decision boundary (un)fairness as a tractable proxy to the rule: the covariance between the sensitive attributes and the (signed) distance between the subjects’ feature vectors and the decision boundary of the classifier.”, Page 4 Section 3.1, “we account for disparate treatment, by removing the sensitive features from the decision making process and, for disparate impact, by adding fairness constraints during the training process of the classifier.”
Zafar teaches fairness constraints (e.g., covariance constraints) during classifier training. The constraints correspond to the debiasing term that modifies the learned decision boundary. Training the SVM using the covariance constraints to reduce bias becomes a latent bias reduced SVM inference model for future predictions/inferences.),
wherein a first amount of latent bias exhibited in first inferences generated by the latent bias reduced support vector machine-based inference model is lower than a second amount of latent bias exhibited in second inferences generated by the support vector machine-based inference model (Page 5 Section 4.1, “We compare these boundaries against the unconstrained decision boundary (solid line)…This movement of the decision boundaries shows that our fairness constraints are successfully undoing (albeit in a highly controlled set ting) the rotations we used to induce disparate impact in the dataset.”, Page 4 Section 3.2, “In this formulation, c trades off fairness and accuracy, such that as we de crease c towards zero, the resulting classifier will satisfy a larger p%-rule but will potentially suffer from a larger loss in accuracy.”
Zafar demonstrates that adding fairness constraints shifts the decision boundary to reduce bias. The comparison between constrained and unconstrained classifiers corresponds to the reduced bias in outputs.),
obtaining the inference using the latent bias reduced support vector machine-based inference model (Page 4 Section 3.2, “To this end, we find the decision boundary parameters θ by minimizing the corresponding loss function over the training set under fairness constraints”, Page 4 Section 3.2, “…the constrained optimization problem (4), can also be written as a regularized optimization problem by making use of its dual form, in which the fairness constraints are moved to the objective and the corresponding Lagrange multipliers act as regularizers.”
Zafar describes applying the trained classifier to unseen inputs to generate predictions. The inference is generated by the trained classifier on test inputs. The classifier predicts a class label for each test feature vector.),
the inference being one of the first inferences generated by the latent bias reduced support vector machine- based inference model (Page 3 Section 2, “Then, given an un seen feature vector xi from the test set, the classifier predicts
PNG
media_image1.png
17
266
media_image1.png
Greyscale
otherwise…”Pages 6 and 7 Section 4.2, “Fig. 2c summarizes the results: the top panel shows average accuracy and the bottom panel the percentage of protected (dashed lines) and non-protected (solid lines) users in positive class against the average p% rule, as computed on test sets.”
Zafar discloses applying a trained (bias-aware) classification model to unseen input data to generate a prediction (inference). Since the model incorporates fairness or bias-mitigation during training, the resulting prediction is produced by a bias-reduced inference model.);
based on the occurrence: obtaining the latent bias reduced support vector machine-based inference model (Page 4 Section 3.2, “To this end, we find the decision boundary parameters θ by minimizing the corresponding loss function over the training set under fairness constraints”, Introduction of Zafar, “In this work, our goal is to design classifiers— specifically, convex margin-based classifiers like logistic regression and support vector machines (SVMs)—that avoid both disparate treatment and disparate impact…”, Page 4 Section 3.1, “Appendix A presents the specialization of our formulation for both linear and non-linear SVM classifiers.”, Page 2 Introduction, “we introduce a novel intuitive measure of decision boundary (un)fairness as a tractable proxy to the rule: the covariance between the sensitive attributes and the (signed) distance between the subjects’ feature vectors and the decision boundary of the classifier.”)
Zafar teaches fairness constraints (e.g., covariance constraints) during classifier training. The constraints correspond to the debiasing term that modifies the learned decision boundary. Training the SVM using the covariance constraints to reduce bias becomes a latent bias reduced SVM inference model for future predictions/inferences.)
Zafar appears to not teach [a] method for providing computer implemented services using inference models, the method comprising… identifying an occurrence of a condition that indicates an inference is necessary to provide the computer implemented service… and providing the computer implemented services using the inference… training a support vector machine-based inference model based on a soft margin…and the debiasing term utilizes a Kullback-Leibler (KL) divergence.
Gauci, in the same field of endeavor, teaches [a] method for providing computer implemented services using inference models, the method comprising (Col. 1 Lines 33-41 of Gauci, “Embodiments… for recommending an application based upon a triggering event... A selected prediction model can use contextual information (e.g., collected before or after the event is detected) to identify an application for presenting to a user for easier access, e.g., allowing access on a lock screen.”, Col. 3 Lines 17-21, “embodiments of the present invention can utilize a prediction model to suggest an application in a given context that is likely to be run by a user when a triggering event occurs.”
Gauci teaches using prediction models to suggest applications to users based on context and triggering events. This corresponds to computer implemented services (application suggestions/UI actions) and inference models (prediction model generating likelihoods).)
identifying an occurrence of a condition that indicates an inference is necessary to provide the computer implemented services (Col. 3 Lines 62-66 of Gauci, “At block 102, a triggering event is detected. Not all events that can occur at a device are triggering events. A triggering event can be identified as sufficiently likely to correlate to unique operation of the device.”, Col. 14 Lines 20-27, “the device determines whether the detected event is a triggering event. To determine whether the detected event is a triggering event, the detected event may be compared to a predetermined list of events… If the detected event matches one of the predetermined list of events, then the detected event may be determined to be a triggering event.”
Gauci teaches detecting triggering events that indicate when a prediction/inference should be performed.);
and providing the computer implemented services using the inference (Col. 4 Lines 33-38 of Gauci, “an action is performed in association with the application. In an embodiment, the action may be the providing of a user interface for a user to select to run the application. The user interface may be provided in various ways, such as by displaying on a screen of the device, projecting onto a surface, or providing an audio interface.", Col. 11 Lines 19-22, “If the event manager 320 determines that it is an opportune time for the suggested application to be outputted to the user, the event manager 320 may output an application 322 to a user interface 324.”
Gauci teaches using prediction outputs to trigger UI actions or application suggestions.).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date to combine Zafar’s teaching of incorporating fairness constraints into a SVM training to reduce bias in model predictions with Gauci’s teaching of utilizing prediction models in response to triggering events to provide application services in order to deploy a bias-reduced inference model within an event-driven system to improve the reliability of the computer-implemented service (Paragraphs 21 and 25 of Gauci).
Zafar in view of Gauci appear to not teach training a support vector machine-based inference model based on a soft margin…and the debiasing term utilizes a Kullback-Leibler (KL) divergence;
Wu, in the same field of endeavor, teaches training a support vector machine-based inference model based on a soft margin (Page 3 Section 3.2 of Wu, “The hard maximal margin classifier is an important concept, but it has two problems… To be able to tolerate noise and outliers, we need to take into consideration the positions of more training samples than just those closest to the boundary. This is done generally by introducing slack variables and soft margin classifier… The soft margin classifier is typically the solution that minimizes the regularized norm of
PNG
media_image2.png
22
148
media_image2.png
Greyscale
… The primal optimization problem of maximal weighted soft margin classifier can thus be formulated as…
PNG
media_image3.png
45
258
media_image3.png
Greyscale
”
Wu teaches soft margin SVM training using slack variables and optimization of regularized objective in order to tolerate noise and outliers in training data.)
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date to combine Zafar in view of Gauci’s teachings with Wu’s teaching of soft-margin SVM training using slack variables and regularized optimization in order to handle noisy and non-separable data to implement the reduced bias SVM model (Section 3.2 of Wu).
Zafar in view of Gauci in further view of Wu appear to not teach and the debiasing term utilizes a Kullback-Leibler (KL) divergence…
Buyl, in the same field of endeavor, teaches and the debiasing term utilizes a Kullback-Leibler (KL) divergence (Page 2 Introduction of Buyl, “…we then propose the KL-divergence between a (possibly biased) fitted probabilistic network model and its fair I-projection as a generic fairness regularizer, to be minimized in combination with the usual cost function for the network model.”, Page 8 Section 4.1, “We propose to add the KL divergence DKL(hF || h) as an extra loss term LF. The overall objective function L to find h is thus:
PNG
media_image4.png
52
235
media_image4.png
Greyscale
with γ a hyperparameter that controls the strength of the loss term.”
Buyl teaches the KL-divergence as a fairness regularizer added to the loss function for the machine learning model.);
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date to combine Zafar in view of Gauci in further view of Wu’s teaching with Buyl’s teaching of incorporating Kullback-Leibler divergence as a regularization term in an objective function of a machine learning model in order to improve control over bias during model training (Section 4.1 of Buyl).).
Regarding claim 2,
Zafar does not teach wherein obtaining the latent bias reduced support vector machine-based inference model comprises: reading the latent bias reduced support vector machine-based inference model from storage.
Gauci, in the same field of endeavor, teaches wherein obtaining the latent bias reduced support vector machine-based inference model comprises: reading the latent bias reduced support vector machine-based inference model from storage (Col. 10 Lines 41-45 of Gauci, “The prediction engine 302 may be program code stored on a memory device. In embodiments, the prediction engine 302 includes one or more prediction models.”, Col. 14 Lines 27-30, “At block 506, the device selects a prediction model, e.g., one of the prediction models 1 through N in FIG. 4.”
Gauci teaches prediction models stored in memory and selected when needed.).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date to combine Zafar’s teaching of incorporating fairness constraints into a SVM training to reduce bias in model predictions with Gauci’s teaching of utilizing prediction models in response to triggering events to provide application services in order to deploy a bias-reduced inference model within an event-driven system to improve the reliability of the computer-implemented service (Paragraphs 21 and 25 of Gauci).
Regarding claim 3,
Zafar teaches wherein the support vector machine-based inference model is further trained using training data (Page 3 Section 2, “This task is achieved by utilizing a training set, {(xi,yi)}N i=1, to construct a mapping that works well on an unseen test set.” Page 6 Section 4.2, “we repeatedly split each dataset into a train (70%) and test (30%) set 5 times and report the average statistics for accuracy and fairness.”
Zafar uses training datasets of labeled feature vectors to learn classifier parameters.)
Regarding claim 4,
Zafar teaches the training data comprises: records, and each of the records comprises: at least one feature value; at least one label value associated with the at least one feature value; and at least one bias feature value associated with the at least one feature value (Page 3 Section 2, “If class labels in the training set are correlated with one or more sensitive attributes {zi}N i=1 (e.g., gender, race), the percentage of users with a certain sensitive attribute having dθ∗(xi) ≥ 0 may differ dramatically from the percentage of users without this sensitive at tribute value having dθ∗(xi) ≥ 0 (i.e., the classifier may suffer from disparate impact).”, Page 6 Section 4.2, “The Adult dataset contains a total of 45,222 subjects, each with 14 features (e.g., age, educational level) and a binary label, which indicates whether a subject’s incomes is above (positive class) or below (negative class) 50K USD. For this dataset, we consider gender and race, respectively, as binary and non-binary (polyvalent) sensitive attributes.”
Zafar describes datasets containing features, labels, and sensitive attributes (bias features). Zafar’s training data consists of subjects (records), each having non-sensitive features such as age and education level (feature value), a binary class label such as income classification (label value), and sensitive attributes such as gender or race (bias feature value).).
Regarding claim 5,
Zafar teaches wherein training the support vector machine-based inference model to obtain the latent bias reduced support vector machine-based inference model comprises: obtaining… based in part on the debiasing term … a decision boundary (Page 4 Section 3.2, “To this end, we find the decision boundary parameters θ by minimizing the corresponding loss function over the training set under fairness constraints”, Page 4 Section 3.2, “…the constrained optimization problem (4), can also be written as a regularized optimization problem by making use of its dual form, in which the fairness constraints are moved to the objective and the corresponding Lagrange multipliers act as regularizers.”
Zafar defines optimization of decision boundary parameters via a loss function with fairness constraints.).
Zafar in view of Gauci appears to not teach an objective function based in part on… the soft margin, a decision boundary.
Wu, in the same field of endeavor, teaches obtaining, based on the training data and an objective function based in part on… the soft margin, a decision boundary (Page 3 Section 3.2 of Wu, “The hard maximal margin classifier is an important concept, but it has two problems… To be able to tolerate noise and outliers, we need to take into consideration the positions of more training samples than just those closest to the boundary. This is done generally by introducing slack variables and soft margin classifier… The soft margin classifier is typically the solution that minimizes the regularized norm of
PNG
media_image2.png
22
148
media_image2.png
Greyscale
… The primal optimization problem of maximal weighted soft margin classifier can thus be formulated as…
PNG
media_image3.png
45
258
media_image3.png
Greyscale
”
Wu teaches soft margin SVM training using slack variables and optimization of regularized objective in order to tolerate noise and outliers in training data.).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date to combine Zafar in view of Gauci’s teachings with Wu’s teaching of soft-margin SVM training using slack variables and regularized optimization in order to handle noisy and non-separable data to implement the reduced bias SVM model (Section 3.2 of Wu).
Regarding claim 6,
Zafar teaches the debiasing term provides an incentive for a uniform distribution of the records with respect to one or more bias features across the decision boundary in the objective function (Page 4 Section 3.1, “if a decision boundary satisfies the “100%-rule”, i.e., P(dθ(x) ≥ 0|z = 0) = P(dθ(x) ≥ 0|z = 1), (3) then the (empirical) covariance will be approximately zero for a sufficiently large training set.”, Page 4 Section 3.1, “Our measure of decision boundary (un)fairness is defined as the covariance between the users’ sensitive attributes, {zi}N i=1, and the signed distance from the users’ feature vectors to the decision boundary, {dθ(xi)}N i=1, i.e.:”
Zafar defines fairness as the covariance between sensitive attributes and the signed distance to the decision boundary and imposes a constraint to reduce this covariance. When the covariance approaches zero, the probabilities of classification outcomes for different sensitive groups become equal, corresponding to a uniform distribution across the decision boundary.).
Regarding claim 7,
Zafar teaches the objective function comprises a weight that scales a level of the incentive for the uniform distribution of the records (Page 4 Section 3.2, “where c is the covariance threshold, which specifies an upper bound on the covariance between each sensitive attribute and the signed distance from the feature vectors to the decision boundary. In this formulation, c trades off fairness and accuracy, such that as we de crease c towards zero, the resulting classifier will satisfy a larger p%-rule but will potentially suffer from a larger loss in accuracy.”, Page 4 Section 3.2, “…in which the fairness constraints are moved to the objective and the corresponding Lagrange multipliers act as regularizers.”
Zafar introduces a covariance threshold parameter c and describes transforming the constraint into the objective using Lagrange multipliers, which act as regularization weights. The parameter determines how strongly the covariance term influences the optimization, thus scales the level of fairness.).
Regarding claim 8,
Zafar teaches training a support vector machine-based inference model based on… a debiasing term to obtain a latent bias reduced support vector machine-based inference model (Page 1 Introduction of Zafar, “In this work, our goal is to design classifiers— specifically, convex margin-based classifiers like logistic regression and support vector machines (SVMs)—that avoid both disparate treatment and disparate impact…”, Page 4 Section 3.1, “Appendix A presents the specialization of our formulation for both linear and non-linear SVM classifiers.”, Page 2 Introduction, “we introduce a novel intuitive measure of decision boundary (un)fairness as a tractable proxy to the rule: the covariance between the sensitive attributes and the (signed) distance between the subjects’ feature vectors and the decision boundary of the classifier.”, Page 4 Section 3.1, “we account for disparate treatment, by removing the sensitive features from the decision making process and, for disparate impact, by adding fairness constraints during the training process of the classifier.”
Zafar teaches fairness constraints (e.g., covariance constraints) during classifier training. The constraints correspond to the debiasing term that modifies the learned decision boundary. Training the SVM using the covariance constraints to reduce bias becomes a latent bias reduced SVM inference model for future predictions/inferences.)
wherein a first amount of latent bias exhibited in first inferences generated by the latent bias reduced support vector machine-based inference model is lower than a second amount of latent bias exhibited in second inferences generated by the support vector machine-based inference model (Page 5 Section 4.1, “We compare these boundaries against the unconstrained decision boundary (solid line)…This movement of the decision boundaries shows that our fairness constraints are successfully undoing (albeit in a highly controlled set ting) the rotations we used to induce disparate impact in the dataset.”, Page 4 Section 3.2, “In this formulation, c trades off fairness and accuracy, such that as we de crease c towards zero, the resulting classifier will satisfy a larger p%-rule but will potentially suffer from a larger loss in accuracy.”
Zafar demonstrates that adding fairness constraints shifts the decision boundary to reduce bias. The comparison between constrained and unconstrained classifiers corresponds to the reduced bias in outputs.)
obtaining the latent bias reduced support vector machine-based inference model (Page 4 Section 3.2, “To this end, we find the decision boundary parameters θ by minimizing the corresponding loss function over the training set under fairness constraints”, Page 4 Section 3.2, “…the constrained optimization problem (4), can also be written as a regularized optimization problem by making use of its dual form, in which the fairness constraints are moved to the objective and the corresponding Lagrange multipliers act as regularizers.”
Zafar describes applying the trained classifier to unseen inputs to generate predictions. The inference is generated by the trained classifier on test inputs. The classifier predicts a class label for each test feature vector.)
the inference being one of the first inferences generated by the latent bias reduced support vector machine- based inference model (Page 3 Section 2, “Then, given an un seen feature vector xi from the test set, the classifier predicts
PNG
media_image1.png
17
266
media_image1.png
Greyscale
otherwise…”Pages 6 and 7 Section 4.2, “Fig. 2c summarizes the results: the top panel shows average accuracy and the bottom panel the percentage of protected (dashed lines) and non-protected (solid lines) users in positive class against the average p% rule, as computed on test sets.”);
based on the occurrence: obtaining the latent bias reduced support vector machine-based inference model (Page 4 Section 3.2, “To this end, we find the decision boundary parameters θ by minimizing the corresponding loss function over the training set under fairness constraints”, Introduction of Zafar, “In this work, our goal is to design classifiers— specifically, convex margin-based classifiers like logistic regression and support vector machines (SVMs)—that avoid both disparate treatment and disparate impact…”, Page 4 Section 3.1, “Appendix A presents the specialization of our formulation for both linear and non-linear SVM classifiers.”, Page 2 Introduction, “we introduce a novel intuitive measure of decision boundary (un)fairness as a tractable proxy to the rule: the covariance between the sensitive attributes and the (signed) distance between the subjects’ feature vectors and the decision boundary of the classifier.”)
Zafar teaches fairness constraints (e.g., covariance constraints) during classifier training. The constraints correspond to the debiasing term that modifies the learned decision boundary. Training the SVM using the covariance constraints to reduce bias becomes a latent bias reduced SVM inference model for future predictions/inferences.)
Zafar appears to not teach [a] method for providing computer implemented services using inference models, the method comprising… identifying an occurrence of a condition that indicates an inference is necessary to provide the computer implemented service… and providing the computer implemented services using the inference… training a support vector machine-based inference model based on a soft margin…and the debiasing term utilizes a Kullback-Leibler (KL) divergence.
Gauci, in the same field of endeavor, teaches [a] non-transitory machine-readable medium having instructions stored therein, which when executed by a processor, cause the processor to perform operations (Paragraph 123, “One or more processors 818 are configurable to process various data formats for one or more application programs 834 stored on medium 802.”).
identifying an occurrence of a condition that indicates an inference is necessary to provide the computer implemented services (Col. 3 Lines 62-66 of Gauci, “At block 102, a triggering event is detected. Not all events that can occur at a device are triggering events. A triggering event can be identified as sufficiently likely to correlate to unique operation of the device.”, Col. 14 Lines 20-27, “the device determines whether the detected event is a triggering event. To determine whether the detected event is a triggering event, the detected event may be compared to a predetermined list of events… If the detected event matches one of the predetermined list of events, then the detected event may be determined to be a triggering event.”
Gauci teaches detecting triggering events that indicate when a prediction/inference should be performed.);
and providing the computer implemented services using the inference (Col. 4 Lines 33-38 of Gauci, “an action is performed in association with the application. In an embodiment, the action may be the providing of a user interface for a user to select to run the application. The user interface may be provided in various ways, such as by displaying on a screen of the device, projecting onto a surface, or providing an audio interface.", Col. 11 Lines 19-22, “If the event manager 320 determines that it is an opportune time for the suggested application to be outputted to the user, the event manager 320 may output an application 322 to a user interface 324.”
Gauci teaches using prediction outputs to trigger UI actions or application suggestions.).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date to combine Zafar’s teaching of incorporating fairness constraints into a SVM training to reduce bias in model predictions with Gauci’s teaching of utilizing prediction models in response to triggering events to provide application services in order to deploy a bias-reduced inference model within an event-driven system to improve the reliability of the computer-implemented service (Paragraphs 21 and 25 of Gauci).
Zafar in view of Gauci appear to not teach training a support vector machine-based inference model based on a soft margin…and the debiasing term utilizes a Kullback-Leibler (KL) divergence;
Wu, in the same field of endeavor, teaches training a support vector machine-based inference model based on a soft margin (Page 3 Section 3.2 of Wu, “The hard maximal margin classifier is an important concept, but it has two problems… To be able to tolerate noise and outliers, we need to take into consideration the positions of more training samples than just those closest to the boundary. This is done generally by introducing slack variables and soft margin classifier… The soft margin classifier is typically the solution that minimizes the regularized norm of
PNG
media_image2.png
22
148
media_image2.png
Greyscale
… The primal optimization problem of maximal weighted soft margin classifier can thus be formulated as…
PNG
media_image3.png
45
258
media_image3.png
Greyscale
”
Wu teaches soft margin SVM training using slack variables and optimization of regularized objective in order to tolerate noise and outliers in training data.)
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date to combine Zafar in view of Gauci’s teachings with Wu’s teaching of soft-margin SVM training using slack variables and regularized optimization in order to handle noisy and non-separable data to implement the reduced bias SVM model (Section 3.2 of Wu).
Zafar in view of Gauci in further view of Wu appear to not teach and the debiasing term utilizes a Kullback-Leibler (KL) divergence…
Buyl, in the same field of endeavor, teaches and the debiasing term utilizes a Kullback-Leibler (KL) divergence (Page 2 Introduction of Buyl, “…we then propose the KL-divergence between a (possibly biased) fitted probabilistic network model and its fair I-projection as a generic fairness regularizer, to be minimized in combination with the usual cost function for the network model.”, Page 8 Section 4.1, “We propose to add the KL divergence DKL(hF || h) as an extra loss term LF. The overall objective function L to find h is thus:
PNG
media_image4.png
52
235
media_image4.png
Greyscale
with γ a hyperparameter that controls the strength of the loss term.”
Buyl teaches the KL-divergence as a fairness regularizer added to the loss function for the machine learning model.);
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date to combine Zafar in view of Gauci in further view of Wu’s teaching with Buyl’s teaching of incorporating Kullback-Leibler divergence as a regularization term in an objective function of a machine learning model in order to improve control over bias during model training (Section 4.1 of Buyl).).
Claim 9 is an article of manufacture claim that recites similar limitations to method claim 2. Therefore, claim 9 is rejected using the same rationale as claim 2.
Claim 10 is an article of manufacture claim that recites similar limitations to method claim 3. Therefore, claim 10 is rejected using the same rationale as claim 3.
Claim 11 is an article of manufacture claim that recites similar limitations to method claim 4. Therefore, claim 11 is rejected using the same rationale as claim 4.
Claim 12 is an article of manufacture claim that recites similar limitations to method claim 5. Therefore, claim 12 is rejected using the same rationale as claim 5.
Claim 13 is an article of manufacture claim that recites similar limitations to method claim 6. Therefore, claim 13 is rejected using the same rationale as claim 6.
Claim 14 is an article of manufacture claim that recites similar limitations to method claim 7. Therefore, claim 14 is rejected using the same rationale as claim 7.
Regarding claim 15,
Zafar teaches training a support vector machine-based inference model based on… a debiasing term to obtain a latent bias reduced support vector machine-based inference model (Page 1 Introduction of Zafar, “In this work, our goal is to design classifiers— specifically, convex margin-based classifiers like logistic regression and support vector machines (SVMs)—that avoid both disparate treatment and disparate impact…”, Page 4 Section 3.1, “Appendix A presents the specialization of our formulation for both linear and non-linear SVM classifiers.”, Page 2 Introduction, “we introduce a novel intuitive measure of decision boundary (un)fairness as a tractable proxy to the rule: the covariance between the sensitive attributes and the (signed) distance between the subjects’ feature vectors and the decision boundary of the classifier.”, Page 4 Section 3.1, “we account for disparate treatment, by removing the sensitive features from the decision making process and, for disparate impact, by adding fairness constraints during the training process of the classifier.”
Zafar teaches fairness constraints (e.g., covariance constraints) during classifier training. The constraints correspond to the debiasing term that modifies the learned decision boundary. Training the SVM using the covariance constraints to reduce bias becomes a latent bias reduced SVM inference model for future predictions/inferences.)
wherein a first amount of latent bias exhibited in first inferences generated by the latent bias reduced support vector machine-based inference model is lower than a second amount of latent bias exhibited in second inferences generated by the support vector machine-based inference model (Page 5 Section 4.1, “We compare these boundaries against the unconstrained decision boundary (solid line)…This movement of the decision boundaries shows that our fairness constraints are successfully undoing (albeit in a highly controlled set ting) the rotations we used to induce disparate impact in the dataset.”, Page 4 Section 3.2, “In this formulation, c trades off fairness and accuracy, such that as we de crease c towards zero, the resulting classifier will satisfy a larger p%-rule but will potentially suffer from a larger loss in accuracy.”
Zafar demonstrates that adding fairness constraints shifts the decision boundary to reduce bias. The comparison between constrained and unconstrained classifiers corresponds to the reduced bias in outputs.)
obtaining the latent bias reduced support vector machine-based inference model (Page 4 Section 3.2, “To this end, we find the decision boundary parameters θ by minimizing the corresponding loss function over the training set under fairness constraints”, Page 4 Section 3.2, “…the constrained optimization problem (4), can also be written as a regularized optimization problem by making use of its dual form, in which the fairness constraints are moved to the objective and the corresponding Lagrange multipliers act as regularizers.”
Zafar describes applying the trained classifier to unseen inputs to generate predictions. The inference is generated by the trained classifier on test inputs. The classifier predicts a class label for each test feature vector.)
the inference being one of the first inferences generated by the latent bias reduced support vector machine- based inference model (Page 3 Section 2, “Then, given an un seen feature vector xi from the test set, the classifier predicts
PNG
media_image1.png
17
266
media_image1.png
Greyscale
otherwise…”Pages 6 and 7 Section 4.2, “Fig. 2c summarizes the results: the top panel shows average accuracy and the bottom panel the percentage of protected (dashed lines) and non-protected (solid lines) users in positive class against the average p% rule, as computed on test sets.”);
based on the occurrence: obtaining the latent bias reduced support vector machine-based inference model (Page 4 Section 3.2, “To this end, we find the decision boundary parameters θ by minimizing the corresponding loss function over the training set under fairness constraints”, Introduction of Zafar, “In this work, our goal is to design classifiers— specifically, convex margin-based classifiers like logistic regression and support vector machines (SVMs)—that avoid both disparate treatment and disparate impact…”, Page 4 Section 3.1, “Appendix A presents the specialization of our formulation for both linear and non-linear SVM classifiers.”, Page 2 Introduction, “we introduce a novel intuitive measure of decision boundary (un)fairness as a tractable proxy to the rule: the covariance between the sensitive attributes and the (signed) distance between the subjects’ feature vectors and the decision boundary of the classifier.”)
Zafar teaches fairness constraints (e.g., covariance constraints) during classifier training. The constraints correspond to the debiasing term that modifies the learned decision boundary. Training the SVM using the covariance constraints to reduce bias becomes a latent bias reduced SVM inference model for future predictions/inferences.)
Zafar appears to not teach [a] method for providing computer implemented services using inference models, the method comprising… identifying an occurrence of a condition that indicates an inference is necessary to provide the computer implemented service… and providing the computer implemented services using the inference… training a support vector machine-based inference model based on a soft margin…and the debiasing term utilizes a Kullback-Leibler (KL) divergence.
Gauci, in the same field of endeavor, teaches [a] data processing system, comprising: a processor; and a memory coupled to the processor to store instructions, which when executed by the processor, cause the processor to perform operations (Paragraph 123 of Gauci, “One or more processors 818 are configurable to process various data formats for one or more application programs 834 stored on medium 802.”, Paragraph 124 of Gauci, “One or more processors 818 communicate with computer-readable medium 802 via a controller 820. Computer-readable medium 802 can be any device or medium that can store code and/or data for use by one or more processors 818. Medium 802 can include a memory hierarchy, including cache, main memory and secondary memory.”).
identifying an occurrence of a condition that indicates an inference is necessary to provide the computer implemented services (Col. 3 Lines 62-66 of Gauci, “At block 102, a triggering event is detected. Not all events that can occur at a device are triggering events. A triggering event can be identified as sufficiently likely to correlate to unique operation of the device.”, Col. 14 Lines 20-27, “the device determines whether the detected event is a triggering event. To determine whether the detected event is a triggering event, the detected event may be compared to a predetermined list of events… If the detected event matches one of the predetermined list of events, then the detected event may be determined to be a triggering event.”
Gauci teaches detecting triggering events that indicate when a prediction/inference should be performed.);
and providing the computer implemented services using the inference (Col. 4 Lines 33-38 of Gauci, “an action is performed in association with the application. In an embodiment, the action may be the providing of a user interface for a user to select to run the application. The user interface may be provided in various ways, such as by displaying on a screen of the device, projecting onto a surface, or providing an audio interface.", Col. 11 Lines 19-22, “If the event manager 320 determines that it is an opportune time for the suggested application to be outputted to the user, the event manager 320 may output an application 322 to a user interface 324.”
Gauci teaches using prediction outputs to trigger UI actions or application suggestions.).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date to combine Zafar’s teaching of incorporating fairness constraints into a SVM training to reduce bias in model predictions with Gauci’s teaching of utilizing prediction models in response to triggering events to provide application services in order to deploy a bias-reduced inference model within an event-driven system to improve the reliability of the computer-implemented service (Paragraphs 21 and 25 of Gauci).
Zafar in view of Gauci appear to not teach training a support vector machine-based inference model based on a soft margin…and the debiasing term utilizes a Kullback-Leibler (KL) divergence;
Wu, in the same field of endeavor, teaches training a support vector machine-based inference model based on a soft margin (Page 3 Section 3.2 of Wu, “The hard maximal margin classifier is an important concept, but it has two problems… To be able to tolerate noise and outliers, we need to take into consideration the positions of more training samples than just those closest to the boundary. This is done generally by introducing slack variables and soft margin classifier… The soft margin classifier is typically the solution that minimizes the regularized norm of
PNG
media_image2.png
22
148
media_image2.png
Greyscale
… The primal optimization problem of maximal weighted soft margin classifier can thus be formulated as…
PNG
media_image3.png
45
258
media_image3.png
Greyscale
”
Wu teaches soft margin SVM training using slack variables and optimization of regularized objective in order to tolerate noise and outliers in training data.)
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date to combine Zafar in view of Gauci’s teachings with Wu’s teaching of soft-margin SVM training using slack variables and regularized optimization in order to handle noisy and non-separable data to implement the reduced bias SVM model (Section 3.2 of Wu).
Zafar in view of Gauci in further view of Wu appear to not teach and the debiasing term utilizes a Kullback-Leibler (KL) divergence…
Buyl, in the same field of endeavor, teaches and the debiasing term utilizes a Kullback-Leibler (KL) divergence (Page 2 Introduction of Buyl, “…we then propose the KL-divergence between a (possibly biased) fitted probabilistic network model and its fair I-projection as a generic fairness regularizer, to be minimized in combination with the usual cost function for the network model.”, Page 8 Section 4.1, “We propose to add the KL divergence DKL(hF || h) as an extra loss term LF. The overall objective function L to find h is thus:
PNG
media_image4.png
52
235
media_image4.png
Greyscale
with γ a hyperparameter that controls the strength of the loss term.”
Buyl teaches the KL-divergence as a fairness regularizer added to the loss function for the machine learning model.);
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date to combine Zafar in view of Gauci in further view of Wu’s teaching with Buyl’s teaching of incorporating Kullback-Leibler divergence as a regularization term in an objective function of a machine learning model in order to improve control over bias during model training (Section 4.1 of Buyl).).
Claim 16 is a machine claim that recites identical limitations to method claim 3. Therefore, claim 16 is
rejected using the same rationale as claim 3.
Claim 17 is a machine claim that recites similar limitations to method claim 4. Therefore, claim 17 is
rejected using the same rationale as claim 4.
Claim 18 is a machine claim that recites similar limitations to method claim 5. Therefore, claim 18 is
rejected using the same rationale as claim 5.
Claim 19 is a machine claim that recites similar limitations to method claim 6. Therefore, claim 19 is
rejected using the same rationale as claim 6.
Claim 20 is a machine claim that recites similar limitations to method claim 7. Therefore, claim 20 is
rejected using the same rationale as claim 7.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any extension fee pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to MAJD HADDAD whose telephone number is (571) 272-2265. The examiner can normally be reached Mon-Friday 8-5 pm.
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.
/M.M.H./Examiner, Art Unit 2125
/KAMRAN AFSHAR/Supervisory Patent Examiner, Art Unit 2125
Prosecution Insights