CTNF 18/503,166 CTNF 100578 DETAILED ACTION Notice of Pre-AIA or AIA Status 07-03-aia AIA 15-10-aia 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 responsive to the application filed on November 6 th , 2023. Claims 1-20 are pending in the case. Claims 1, 8, and 14 are independent claims. Information Disclosure Statement The information disclosure statement (IDS) submitted on November 6 th , 2023 is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner. Claim Rejections - 35 USC § 112 07-30-02 AIA The following is a quotation of 35 U.S.C. 112(b): (b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention. The following is a quotation of 35 U.S.C. 112 (pre-AIA), second paragraph: The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention. 07-34-01 Claims 6, 13, and 19 rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention. Claim 6, 13, and 19 recite the limitation “ identifying non-trivial regimes where uncertainty incurs no performance loss ”. It is unclear whether the uncertainty is the same uncertainty introduced in claims 5, 12, and 18 respectively, or a new uncertainty. For examination purposes, this limitation has been interpreted as “identifying non-trivial regimes where the uncertainty incurs no performance loss”. Claim Rejections - 35 USC § 101 07-04-01 AIA 07-04 35 U.S.C. 101 reads as follows: Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title. Claims 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. Regarding claim 1: Step 1 : Claim 1 is directed to [a] computer-implemented method , therefore it falls under the statuary category of a process. Step 2A Prong 1 : The claim recites, in part: “ reducing…one or more tasks…to a quadratically constrained quadratic problem (QCQP) ” this limitation is a mathematical concept. “ characterizing…one or more solutions associated with the QCQP, wherein each of the one or more solutions is a loss function value ” this encompasses the mental characterization of observed solutions. Further, this limitation is a mathematical concept. Step 2A Prong 2 : The judicial exception is not integrated into a practical application; the remaining limitations of the claim are as follows: “ learning, by the computation engine, the predictive model having missing or crippled data ”, “ applying, via the processor, a formulated problem of missing or crippled data based learning to the predictive model ”, “ by the computation engine ”, “ via the computation engine ” the limitations are an additional element that amounts to adding the words “apply it” (or an equivalent) with the judicial exception, or merely uses a computer in its ordinary capacity as a tool to perform an existing process. See MPEP § 2106.05(f)(2). “ associated with the predictive model ” the limitation is an additional element that generally links the use of the judicial exception to a particular technological environment or field of use. See MPEP § 2106.05(h). Step 2B : The additional elements, taken individually and in combination, do not provide an inventive concept of significantly more than the abstract idea itself for the reasons set forth in step 2A prong 2 above. Therefore, the claim is ineligible. Regarding claim 2, the rejection of claim 2 is incorporated and further: Step 2A Prong 1 : The claim recites, in part: “ the missing or crippled data is related to one or more sensitive attributes ” a continuation of the abstract idea identified in the parent claim. Step 2A Prong 2 : The claim does not recite any additional limitations, thus does not further recite any additional elements that integrates the judicial exception into a practical application or amount to significantly more. Regarding claim 3, the rejection of claim 2 is incorporated and further: Step 2A Prong 1 : The claim recites, in part: “ the formulated problem is based on an independence of sensitive attribute criterion ” Further, this limitation is a mathematical concept. Step 2A Prong 2 : The claim does not recite any additional limitations, thus does not further recite any additional elements that integrates the judicial exception into a practical application or amount to significantly more. Regarding claim 4, the rejection of claim 2 is incorporated and further: Step 2A Prong 1 : The claim recites, in part: “ each of the one or more tasks comprise: a classification task or a regression task ” a continuation of the abstract idea identified in the parent claim. Step 2A Prong 2 : The claim does not recite any additional limitations, thus does not further recite any additional elements that integrates the judicial exception into a practical application or amount to significantly more. Regarding claim 5, the rejection of claim 2 is incorporated and further: Step 2A Prong 1 : The claim recites, in part: “ in response to uncertainty from a limited amount of labeled data, identifying a contribution of each of the one or more solutions towards an improved performance of the QCQP achieved with unlimited access to labeled sensitive attributes ” this encompasses the mental realization that there is an uncertainty arising from a limited amount of data and, in response, identifying a contribution of observed solutions towards an improved performance of the QCQP achieved with unlimited access to data. Further, this limitation is a mathematical concept. Step 2A Prong 2 : The claim does not recite any additional limitations, thus does not further recite any additional elements that integrates the judicial exception into a practical application or amount to significantly more. Regarding claim 6, the rejection of claim 5 is incorporated and further: Step 2A Prong 1 : The claim recites, in part: “ identifying non-trivial regimes where uncertainty incurs no performance loss…and embodies a strict sensitive attribute inclusion ” this encompasses the mental identification of observed regimes where an observed uncertainty does not incur a performance loss. Step 2A Prong 2 : The judicial exception is not integrated into a practical application; the remaining limitations of the claim are as follows: “ associated with the predictive model ” the limitation is an additional element that generally links the use of the judicial exception to a particular technological environment or field of use. See MPEP § 2106.05(h). Step 2B : The additional elements, taken individually and in combination, do not provide an inventive concept of significantly more than the abstract idea itself for the reasons set forth in step 2A prong 2 above. Therefore, the claim is ineligible. Regarding claim 7, the rejection of claim 1 is incorporated and further: Step 2A Prong 1 : The claim recites, in part: “ a generic bootstrap-based algorithm is utilized by the predictive model for non-Gaussian data ” this limitation is a mathematical concept. Step 2A Prong 2 : The claim does not recite any additional limitations, thus does not further recite any additional elements that integrates the judicial exception into a practical application or amount to significantly more. Regarding claim 8: Step 1 : Claim 8 is directed to [a] computer program product , therefore it falls under the statuary category of a manufacture. Step 2A Prong 1 : The claim recites, in part: “ reducing…one or more tasks…to a quadratically constrained quadratic problem (QCQP) ” this limitation is a mathematical concept. “ characterizing…one or more solutions associated with the QCQP, wherein each of the one or more solutions is a loss function value ” this encompasses the mental characterization of observed solutions. Further, this limitation is a mathematical concept. Step 2A Prong 2 : The judicial exception is not integrated into a practical application; the remaining limitations of the claim are as follows: “ learning, by a computation engine, the predictive model having missing or crippled data ”, “ applying, via the processor, a formulated problem of missing or crippled data based learning to the predictive model ”, “ by the computation engine ”, “ via the computation engine ” the limitations are an additional element that amounts to adding the words “apply it” (or an equivalent) with the judicial exception, or merely uses a computer in its ordinary capacity as a tool to perform an existing process. See MPEP § 2106.05(f)(2). “ associated with the predictive model ” the limitation is an additional element that generally links the use of the judicial exception to a particular technological environment or field of use. See MPEP § 2106.05(h). Step 2B : The additional elements, taken individually and in combination, do not provide an inventive concept of significantly more than the abstract idea itself for the reasons set forth in step 2A prong 2 above. Therefore, the claim is ineligible. Regarding claims 9-13: The rejection of claim 8 is further incorporated, the rejections of claims 2-6 are applicable to claims 9-13, respectively. Regarding claim 14: Step 1 : Claim 14 is directed to [a] computing system , therefore it falls under the statuary category of a machine. Step 2A Prong 1 : The claim recites, in part: “ reducing…one or more tasks…to a quadratically constrained quadratic problem (QCQP) ” this limitation is a mathematical concept. “ characterizing…one or more solutions associated with the QCQP, wherein each of the one or more solutions is a loss function value ” this encompasses the mental characterization of observed solutions. Further, this limitation is a mathematical concept. Step 2A Prong 2 : The judicial exception is not integrated into a practical application; the remaining limitations of the claim are as follows: “ execution of the program instructions by the processor configures the processor to perform a predictive model generating method comprising ”, “ learning, by a computation engine, the predictive model having missing or crippled data ”, “ applying, via the processor, a formulated problem of missing or crippled data based learning to the predictive model ”, “ by the computation engine ”, “ via the computation engine ” the limitations are an additional element that amounts to adding the words “apply it” (or an equivalent) with the judicial exception, or merely uses a computer in its ordinary capacity as a tool to perform an existing process. See MPEP § 2106.05(f)(2). “ a processor ”, “ a computer-readable storage device coupled to the processor ”, “ a computation engine coupled to the processor ”, “ program instructions stored on the computer-readable storage device for execution by the processor via a memory ”, “ associated with the predictive model ” these limitations are an additional element that generally links the use of the judicial exception to a particular technological environment or field of use. See MPEP § 2106.05(h). Step 2B : The additional elements, taken individually and in combination, do not provide an inventive concept of significantly more than the abstract idea itself for the reasons set forth in step 2A prong 2 above. Therefore, the claim is ineligible. Regarding claims 15-20: The rejection of claim 14 is further incorporated, the rejection of claims 2-7 are applicable to claims 15-20, respectively. Claim Rejections - 35 USC § 103 07-20-aia AIA 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. 07-21-aia AIA Claim s 1-4, 8-11, and 14-17 are rejected under 35 U.S.C. § 103 as being unpatentable over Wang et al. (“Robust Optimization for Fairness with Noisy Protected Groups”, Wang et al., 10 Nov 2020) ( hereinafter “Wang”) in view of Park et al. (“General Heuristics for Nonconvex Quadratically Constrained Quadratic Programming”, Park et al., March 22, 2017) ( hereinafter “Park”) . Regarding claim 1: Wang teaches [a] computer-implemented method for generating a predictive model using a computation engine and a processor; the method comprising: learning, by the computation engine, the predictive model having missing or crippled data (Wang, page 2, ¶2 “In this paper, we focus on the problem of training binary classifiers with fairness constraints when only noisy labels, G ^ ∈ {1,..., m ^ }, are available for m true protected groups, G ∈ {1,...,m}, of interest.” Here, noisy data can be considered the crippled data, and in light of the spec, ¶4 “Typically, it is assumed that true sensitive attributes are available for every sample in training, but in reality, labeled sensitive attributes are often missing or noisy.”) ; applying, via the processor (Wang, page 32, section F.1.3, ¶1 “All optimization code was written in Python and TensorFlow 5.” Here, code written in Python and TensorFlow inherently runs on a processor ) , a formulated problem of missing or crippled data based learning to the predictive model (Wang, page 3, section 3, ¶1 “We begin with the training problem for incorporating group-based fairness criteria in a learning setting [27, 32, 17, 2, 14]. Let X ∈ X ⊆ RD be a random variable representing a feature vector, with a random binary label Y ∈ Y = {0,1} and random protected group membership G ∈ g = {1,...,m}. In addition, let G ^ ∈ g ^ = {1,..., m ^ } be a random variable representing the noisy protected group label for each (X,Y), which we assume we have access to during training.” Here, the learning of noisy data can be considered crippled data based learning ) ; …wherein each of the one or more solutions is a loss function value (Wang, page 9, section 3.2, ¶1 “The new problem, formed by linearizing all the nonconvex constraints using the method described above, is convex and can be solved efficiently to produce a new feasible point x (1) with a lower objective value. If we linearize again the problem around x (1) and repeat the procedure, we get a sequence of feasible points with decreasing objective values.” Here, the objective values can be considered the loss values ) . Wang does not teach “ reducing, by the computation engine, one or more tasks associated with the predictive model to a quadratically constrained quadratic problem (QCQP); characterizing, via the computation engine, one or more solutions associated with the QCQP ” However, Park teaches reducing, by the computation engine, one or more tasks associated with the predictive model to a quadratically constrained quadratic problem (QCQP) (Park, page 4, section 1.1, ¶1 “A quadratically constrained quadratic program (QCQP) is an optimization problem that can be written in the following form: PNG media_image1.png 154 1521 media_image1.png Greyscale where x ∈ R n is the optimization variable, and P i ∈ R n x n , q i ∈ R n , r i ∈ R are given problem data, for i = 0,1,..., m .” Here, the problem data can be considered the tasks reduced to the QCQP ) ; and characterizing, via the computation engine, one or more solutions associated with the QCQP (Park, page 33, ¶3 “Since n is small, we can enumerate every one of 2n = 1024 feasible points and verify that x.value is a global solution of the problem.”) Wang and Park are analogous art because both references concern methods for problem solving and optimization. Accordingly, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Wang’s sensitive attribute and fairness prediction to incorporate the quadratically constrained quadratic problems (QCQP) taught by Park. The motivation for doing so would have been to provide an easily accessible interface to various heuristics for NP-hard QCQP, as stated in Park, page 33, section 6, ¶2 “The primary goal of the package, instead, is to provide an easily accessible interface to various heuristics for NP-hard QCQPs that do not have specialized methods”. Regarding claim 2: Wang in view of Park teaches [t]he method of Claim 1, wherein the missing or crippled data is related to one or more sensitive attributes (Wang, page 2, ¶1 “One important practical question is whether or not these fairness notions can be reliably measured or enforced if the protected group information is noisy, missing, or unreliable. For example, survey participants may be incentivized to obfuscate their responses for fear of disclosure or discrimination, or may be subject to other forms of response bias. Social desirability response bias may affect participants’ answers regarding religion, political affiliation, or sexual orientation [40].”) . Regarding claim 3: Wang in view of Park teaches [t]he method of Claim 2, wherein the formulated problem is based on an independence of sensitive attribute criterion (Wang, page 4, ¶2, “The constraint functions gj(θ) = E[h(θ,X,Y )|G = j] for j ∈ G, where h(θ,X,Y ) is the target fairness metric, e.g. h(θ,X,Y ) = 1 φ(X;θ) > 0 − E[1 φ(X;θ) > 0 ] when equalizing positive rates for the demographic parity [21] criterion (see [14] for more examples).” Here, the demographic parity can be considered the independence of sensitive attribute criterion in light of the spec, ¶88 “The independence criterion, also called demographic parity, demands that u ⊥ e, meaning that predictions should not reveal any information about sensitive attributes.”) . Regarding claim 4: Wang in view of Park teaches [t]he method of Claim 1, wherein each of the one or more tasks comprise: a classification task or a regression task (Wang, page 4, ¶1 “Let φ(X;θ) represent a binary classifier with parameters θ ∈ Θ where φ(X;θ) > 0indicates a positive classification.”) . Regarding claim 8: Wang teaches [a] computer program product for generating a predictive model, the computer program product comprising a computer readable storage medium having program instructions embodied therewith (Wang, page 32, section F.1.3, ¶1 “All optimization code was written in Python and TensorFlow 5.”) , the program instructions executable by a processor to cause the processor to perform: learning, by a computation engine, the predictive model having missing or crippled data (Wang, page 2, ¶2 “In this paper, we focus on the problem of training binary classifiers with fairness constraints when only noisy labels, G ^ ∈ {1,..., m ^ }, are available for m true protected groups, G ∈ {1,...,m}, of interest.” Here, noisy data can be considered the crippled data, and in light of the spec, ¶4 “Typically, it is assumed that true sensitive attributes are available for every sample in training, but in reality, labeled sensitive attributes are often missing or noisy.”) ; applying, via the processor (Wang, page 32, section F.1.3, ¶1 “All optimization code was written in Python and TensorFlow 5.” Here, code written in Python and TensorFlow inherently runs on a processor ) , a formulated problem of missing or crippled data based learning to the predictive model (Wang, page 3, section 3, ¶1 “We begin with the training problem for incorporating group-based fairness criteria in a learning setting [27, 32, 17, 2, 14]. Let X ∈ X ⊆ RD be a random variable representing a feature vector, with a random binary label Y ∈ Y = {0,1} and random protected group membership G ∈ g = {1,...,m}. In addition, let G ^ ∈ g ^ = {1,..., m ^ } be a random variable representing the noisy protected group label for each (X,Y), which we assume we have access to during training.” Here, the learning of noisy data can be considered crippled data based learning ) ; …wherein each of the one or more solutions is a loss function value (Wang, page 9, section 3.2, ¶1 “The new problem, formed by linearizing all the nonconvex constraints using the method described above, is convex and can be solved efficiently to produce a new feasible point x (1) with a lower objective value. If we linearize again the problem around x (1) and repeat the procedure, we get a sequence of feasible points with decreasing objective values.” Here, the objective values can be considered the loss values ) . Wang does not teach “ reducing, by the computation engine, one or more tasks associated with the predictive model to a quadratically constrained quadratic problem (QCQP); characterizing, via the computation engine, one or more solutions associated with the QCQP ” However, Park teaches reducing, by the computation engine, one or more tasks associated with the predictive model to a quadratically constrained quadratic problem (QCQP) (Park, page 4, section 1.1, ¶1 “A quadratically constrained quadratic program (QCQP) is an optimization problem that can be written in the following form: PNG media_image1.png 154 1521 media_image1.png Greyscale where x ∈ R n is the optimization variable, and P i ∈ R n x n , q i ∈ R n , r i ∈ R are given problem data, for i = 0,1,..., m .” Here, the problem data can be considered the tasks reduced to the QCQP ) ; and characterizing, via the computation engine, one or more solutions associated with the QCQP (Park, page 33, ¶3 “Since n is small, we can enumerate every one of 2n = 1024 feasible points and verify that x.value is a global solution of the problem.”) Wang and Park are analogous art because both references concern methods for problem solving and optimization. Accordingly, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Wang’s sensitive attribute and fairness prediction to incorporate the quadratically constrained quadratic problems (QCQP) taught by Park. The motivation for doing so would have been to provide an easily accessible interface to various heuristics for NP-hard QCQP, as stated in Park, page 33, section 6, ¶2 “The primary goal of the package, instead, is to provide an easily accessible interface to various heuristics for NP-hard QCQPs that do not have specialized methods”. Regarding claims 9-11: Claims 9-11 are rejected under the same rationale as claims 2-4, respectively. Regarding claim 14: Wang teaches [a] computing system comprising: a processor (Wang, page 32, section F.1.3, ¶1 “All optimization code was written in Python and TensorFlow 5.” Here, code written in Python and TensorFlow inherently runs on a processor ) ; a computer-readable storage device coupled to the processor; a computation engine coupled to the processor; program instructions stored on the computer-readable storage device for execution by the processor via a memory (Wang, page 32, section F.1.3, ¶1 “All optimization code was written in Python and TensorFlow 5.” Here, code written in Python and TensorFlow inherently runs on a processor ) , wherein execution of the program instructions by the processor configures the processor to perform a predictive model generating method comprising: learning, by a computation engine, the predictive model having missing or crippled data (Wang, page 2, ¶2 “In this paper, we focus on the problem of training binary classifiers with fairness constraints when only noisy labels, G ^ ∈ {1,..., m ^ }, are available for m true protected groups, G ∈ {1,...,m}, of interest.” Here, noisy data can be considered the crippled data, and in light of the spec, ¶4 “Typically, it is assumed that true sensitive attributes are available for every sample in training, but in reality, labeled sensitive attributes are often missing or noisy.”) ; applying, via the processor (Wang, page 32, section F.1.3, ¶1 “All optimization code was written in Python and TensorFlow 5.” Here, code written in Python and TensorFlow inherently runs on a processor ) , a formulated problem of missing or crippled data based learning to the predictive model (Wang, page 3, section 3, ¶1 “We begin with the training problem for incorporating group-based fairness criteria in a learning setting [27, 32, 17, 2, 14]. Let X ∈ X ⊆ RD be a random variable representing a feature vector, with a random binary label Y ∈ Y = {0,1} and random protected group membership G ∈ g = {1,...,m}. In addition, let G ^ ∈ g ^ = {1,..., m ^ } be a random variable representing the noisy protected group label for each (X,Y), which we assume we have access to during training.” Here, the learning of noisy data can be considered crippled data based learning ) ; …wherein each of the one or more solutions is a loss function value (Wang, page 9, section 3.2, ¶1 “The new problem, formed by linearizing all the nonconvex constraints using the method described above, is convex and can be solved efficiently to produce a new feasible point x (1) with a lower objective value. If we linearize again the problem around x (1) and repeat the procedure, we get a sequence of feasible points with decreasing objective values.” Here, the objective values can be considered the loss values ) . Wang does not teach “ reducing, by the computation engine, one or more tasks associated with the predictive model to a quadratically constrained quadratic problem (QCQP); characterizing, via the computation engine, one or more solutions associated with the QCQP ” However, Park teaches reducing, by the computation engine, one or more tasks associated with the predictive model to a quadratically constrained quadratic problem (QCQP) (Park, page 4, section 1.1, ¶1 “A quadratically constrained quadratic program (QCQP) is an optimization problem that can be written in the following form: PNG media_image1.png 154 1521 media_image1.png Greyscale where x ∈ R n is the optimization variable, and P i ∈ R n x n , q i ∈ R n , r i ∈ R are given problem data, for i = 0,1,..., m .” Here, the problem data can be considered the tasks reduced to the QCQP ) ; and characterizing, via the computation engine, one or more solutions associated with the QCQP (Park, page 33, ¶3 “Since n is small, we can enumerate every one of 2n = 1024 feasible points and verify that x.value is a global solution of the problem.”) Wang and Park are analogous art because both references concern methods for problem solving and optimization. Accordingly, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Wang’s sensitive attribute and fairness prediction to incorporate the quadratically constrained quadratic problems (QCQP) taught by Park. The motivation for doing so would have been to provide an easily accessible interface to various heuristics for NP-hard QCQP, as stated in Park, page 33, section 6, ¶2 “The primary goal of the package, instead, is to provide an easily accessible interface to various heuristics for NP-hard QCQPs that do not have specialized methods”. Regarding claims 15-17: Claims 15-17 are rejected under the same rationale as claims 2-4, respectively . 07-21-aia AIA Claim s 5, 12, and 18 are rejected under 35 U.S.C. § 103 as being unpatentable over Wang in view of Park in view of Kallus et al. (“Assessing Algorithmic Fairness with Unobserved Protected Class Using Data Combination”, Kallus et al., 16 Jun 2020) ( hereinafter “Kallus”) in further view of Zhang et al. (“Assessing Fairness in the Presence of Missing Data”, Zhang et al., 7 Dec 2021) ( hereinafter “Zhang”) . Regarding claim 5: Wang in view of Park teaches [t]he method of Claim 1, Wang in view of Park does not teach “ further comprising, in response to uncertainty from a limited amount of labeled data ” However, Kallus teaches further comprising, in response to uncertainty from a limited amount of labeled data (Kallus, page 19, ¶1-2 “We further discuss the consistency of our estimates and inferential procedures for constructing confidence intervals that characterize the additional uncertainty due to finite-sample variability. In this section, instead of assuming access to the population-level marginal distributions, we assume we are given finite-sample datasets.”) Wang in view of Park and Kallus are analogous art because both references concern methods for fairness with sensitive data. Accordingly, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Wang/Park’s fairness and sensitive attribute prediction system to incorporate the confidence intervals taught by Kallus. The motivation for doing so would have been to support credible, principled conclusions about disparities, as stated in Kallus, page 4, ¶1 “…our proposed methods can support credible, principled conclusions about disparities. In particular, these quantify the fundamental ambiguity in disparities and the value of more informative proxies or assumptions, especially if our partial identification sets are large; or, if the sets are small, they provide a statistical test certifying the presence of disparities independent of further untestable assumptions.” Wang in view of Park in further view of Kallus does not teach “ identifying a contribution of each of the one or more solutions towards an improved performance of the QCQP achieved with unlimited access to labeled sensitive attributes ” However, Zhang teaches identifying a contribution of each of the one or more solutions towards an improved performance of the QCQP achieved with unlimited access to labeled sensitive attributes (Zhang, page 5, section 3.1, ¶3 “In the first term of the upper bound, ωDSa is an approximation to the complete data domain in sensitive group a, and dTV(DTa ||ωDSa ) can be viewed as the approximation error. It follows that a less accurate approximation of the complete data domain would lead to a looser upper bound on fairness estimation error.” Here, the complete data domain can be considered the unlimited access to labeled sensitive attributes and a more accurate approximation of the complete data domain can be considered the improved performance ) . Wang in view of Park in further view of Kallus and Zhang are analogous art because both references concern methods for fairness with attribute prediction and sensitive data. Accordingly, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Wang/Park/Kallus’ fairness and sensitive attribute prediction system to incorporate the access taught taught by Zhang. The motivation for doing so would have been to have a fairness guarantee, as stated in Zhang, page 1, Abstract “To fill this significant gap, we study the problem of estimating fairness in the complete data domain for an arbitrary model evaluated merely using complete cases. We provide upper and lower bounds on the fairness estimation error and conduct numerical experiments to assess our theoretical results. Our work provides the first known theoretical results on fairness guarantee in analysis of incomplete data.” Regarding claims 12 and 18: Claims 12 and 18 are rejected under the same rationale as claim 5 . 07-21-aia AIA Claim s 6, 13, and 19 are rejected under 35 U.S.C. § 103 as being unpatentable over Wang in view of Park in view of Kallus in view of Zhang in further view of Menon et al. (“The cost of fairness in classification”, Menon et al., 25 May 2017) ( hereinafter “Menon”) . Regarding claim 6: Wang in view of Park in view of Kallus in further view of Zhang teaches [t]he method of Claim 5 Wang in view of Park in view of Kallus in further view of Zhang does not teach “ wherein the identifying further comprises identifying non-trivial regimes where uncertainty incurs no performance loss associated with the predictive model and embodies a strict sensitive attribute inclusion ” However, Menon teaches wherein the identifying further comprises identifying non-trivial regimes where uncertainty incurs no performance loss associated with the predictive model (Menon, page 11, proposition 8, ¶2 “Proposition 8 specifies how much fairness we can ask for without paying any performance penalty.” Here, the identification of fairness levels that do not have a performance penalty can be considered uncertainty that incurs no performance loss ) and embodies a strict sensitive attribute inclusion (Menon, page 5, section 5.2, ¶2 “This is a simple consequence of Proposition 3, as we can simply consider one of the features of X to be perfectly predictive of the sensitive feature, which makes ηDP(x, y - ) ∈ {0,1}.” Here, the inclusion of sensitive features in the prediction can be considered to embody a strict sensitive attribute inclusion ) . Wang in view of Park in view of Kallus in further view of Zhang and Menon are analogous art because both references concern methods for fairness with machine learning. Accordingly, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Wang/Park/Kallus/Zhang’s fairness and sensitive attribute prediction system to incorporate the lack of performance loss and attribute inclusion taught by Menon. The motivation for doing so would have been to have an abstract, generic treatment of the problem, as stated in Menon “Underpinning our analysis is a general framework casting the fairness-aware learning problem as one of minimising the difference of two statistical risks (§2.2), which allows for an abstract, generic treatment of the problem.” Regarding claims 13 and 19: Claims 13 and 19 are rejected under the same rationale as claim 6 . 07-21-aia AIA Claim s 7 and 20 are rejected under 35 U.S.C. § 103 as being unpatentable over Wang in view of Park in view of Johnston et al. (“A bootstrap approach is a superior statistical method for the comparison of non-normal data with differing variances”, Johnston et al., 22 December 2020) ( hereinafter “Johnston”) . Regarding claim 7: Wang in view of Park teaches [t]he method of Claim 1, Wang in view of Park does not teach “ wherein a generic bootstrap-based algorithm is utilized by the predictive model for non-Gaussian data ” However, Johnston teaches wherein a generic bootstrap-based algorithm is utilized by the predictive model for non-Gaussian data (Johnston, page 1, col 1, ¶1 “When data are not normal, the adage is to use a nonparametric test for this analysis: the most common being the Mann–Whitney–Wilcoxon (MWW) test. Here, we explore conditions in which the MWW test is unsuitable and propose the use of a bootstrap approach instead.” Here, not normal data can be considered the non-gaussian data ) . Wang in view of Park and Johnston are analogous art because both references concern methods for data fairness. Accordingly, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Wang/Park’s fairness and sensitive attribute prediction system to incorporate the bootstrapping taught by Johnston. The motivation for doing so would have been to avoid having to use a known distribution on non-gaussian data, as stated in Johnston, page 1, col 2, ¶2 “Bootstrapping techniques can be used to generate a null distribution de novo from the observed data already collected if the samples are independent. This removes the requirement of using a known distribution.” Regarding claim 20: Claim 20 is rejected under the same rationale as claim 7 . Conclusion 07-96 AIA The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Shah et al. (“Selective Regression Under Fairness Criteria”, Shah et al., 14 Jul 2022) discloses a fairness criteria for selective regression requiring the performance of every subgroup to improve with a decrease in coverage and two approaches to mitigate the performance disparity across subgroups: (a) by regularizing an upper bound of conditional mutual information under a Gaussian assumption and (b) by regularizing a contrastive loss for conditional mean and conditional variance prediction. Mary et al. (“Fairness-Aware Learning for Continuous Attributes and Treatments”, Mary et al., 2019) discloses ensuring that the outcome of a classifier is not biased towards certain values of sensitive variables such as age, race or gender. As common fairness metrics can be expressed as measures of (conditional) independence between variables, we propose to use the Rényi maximum correlation coefficient to generalize fairness measurement to continuous variables. Komiyama et al. (“Nonconvex Optimization for Regression with Fairness Constraints”, Komiyama et al., 2018) discloses using QCQP and the following aspects: (i) The exact control on the fairness level as a constraint, (ii) a capability of dealing with numeric and multiple s and (iii) an extension that captures non-linear interaction between sensitive and non-sensitive attributes. We consider this result as a first step that controls the nonconvexity that naturally appears in considering fairness related constraints. Any inquiry concerning this communication or earlier communications from the examiner should be directed to JACOB Z SUSSMAN MOSS whose telephone number is (571) 272-1579. The examiner can normally be reached Monday - Friday, 9 a.m. - 5 p.m. 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, Kakali Chaki can be reached on (571) 272-3719. 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. /J.S.M./Examiner, Art Unit 2122 /KAKALI CHAKI/Supervisory Patent Examiner, Art Unit 2122 Application/Control Number: 18/503,166 Page 2 Art Unit: 2122 Application/Control Number: 18/503,166 Page 3 Art Unit: 2122 Application/Control Number: 18/503,166 Page 4 Art Unit: 2122 Application/Control Number: 18/503,166 Page 5 Art Unit: 2122 Application/Control Number: 18/503,166 Page 6 Art Unit: 2122 Application/Control Number: 18/503,166 Page 7 Art Unit: 2122 Application/Control Number: 18/503,166 Page 8 Art Unit: 2122 Application/Control Number: 18/503,166 Page 9 Art Unit: 2122 Application/Control Number: 18/503,166 Page 10 Art Unit: 2122 Application/Control Number: 18/503,166 Page 11 Art Unit: 2122 Application/Control Number: 18/503,166 Page 12 Art Unit: 2122 Application/Control Number: 18/503,166 Page 13 Art Unit: 2122 Application/Control Number: 18/503,166 Page 14 Art Unit: 2122 Application/Control Number: 18/503,166 Page 15 Art Unit: 2122 Application/Control Number: 18/503,166 Page 16 Art Unit: 2122 Application/Control Number: 18/503,166 Page 17 Art Unit: 2122 Application/Control Number: 18/503,166 Page 18 Art Unit: 2122 Application/Control Number: 18/503,166 Page 19 Art Unit: 2122 Application/Control Number: 18/503,166 Page 20 Art Unit: 2122 Application/Control Number: 18/503,166 Page 21 Art Unit: 2122
Prosecution Insights