DETAILED ACTION
This non-final office action is responsive to application 18/498,091 as submitted 31 Oct. 2023.
Claim status is currently pending and under examination for claims 1-20 of which independent claims are 1, 11 and 16.
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Information Disclosure Statement
As required by MPEP 609(c), the applicant’s submissions of the Information Disclosure Statement dated 10/31/2023 is acknowledged by the examiner and the cited references have been considered in the examination of the claims now pending. As required by MPEP 609 C(2), a copy of the PTOL-1449 initialed and dated by the examiner is attached to the instant office action.
Claim Rejections - 35 USC § 112
Claims 7, 13 and 18 are rejected under 35 U.S.C. 112(b), as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor, regards as the invention. Particularly, claims recite the limitation "the learning model" which lacks proper antecedent basis. In this case, ‘the learning model’ could be either of “gaussian mixture model” from claim 6 or “generative adversarial network (GAN)” from claim 1 under the broadest reasonable interpretation. Therefore, it is unclear which model is ‘the’ learning model, and will be interpreted as either for purposes of examination. Accordingly, claims are rejected under 35 U.S.C. 112(b) as indefinite for lacking antecedent basis.
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. In determining whether the claims are subject matter eligible, the examiner applies guidance set forth under MPEP 2106.
Step 1: Is the claim to a process, machine, manufacture, or composition of matter? Yes—all claims fall within one of the four statutory categories: claims 1-10 are a method/process, claims 11-15 are a system/machine, and claims 16-20 are a computer program product/article of manufacture. Thus, all claims are to identified statutory categories and the analysis should proceed per MPEP 2106.03.
Step 2A, prong one: Does the claim recite an abstract idea, law of nature or natural phenomenon? Yes—the claims, under the broadest reasonable interpretation, recites an abstract idea. In this case, claims fall within the enumerated grouping of abstract idea being “Mental Processes” and/or “Mathematical Concepts”, but for the recitation of generic computer components. In particular, claims recite:
“identifying a set of training data comprising at least one independent variable and a dependent variable” (mental observation)
“analyzing the set of training data to identify correlations between the at least one independent variable and the dependent variable” (mental evaluation or math relationship)
“identifying the at least one correlation between a first independent variable of the one or more independent variables and the dependent variables” (mental observation, evaluation or mathematical relationships)
“calculating a fairness score for each identified correlation, including for the first independent variable, against the dependent variable” (mathematical calculation or mental evaluation, judgment)
“creating, based on the analyzing, a fairness profile for the set of training data” (mental judgment or math analysis)
“fairness score being below a fairness threshold… increase the fairness score for the first independent variable with above a fairness threshold” (math calculation, relationship or mental judgment, evaluation)
Focus of the claim concern fairness, e.g. bias. The fairness is calculated based on correlations between variables and subject to fairness threshold. Analyzing fairness according to the claimed limitations are functions which, under the broadest reasonable interpretation, may be carried out as part of a human mental process and/or mathematical evaluation. For example, prejudice or partiality measured by ranking with cutoff or templates to represent diversity of datasets that may be observed to include sensitive data like race or gender. When read in light of the instant specification, e.g. [0074] “80% rule is commonly used… can be changed based on each use-case” As such, the claims may recite mental processes and/or mathematical concepts which are enumerated as abstract ideas under MPEP 2106.04(a)(2).
Step 2A, prong two: Does the claim recite additional elements that integrate the judicial exception into a practical application? No—a practical application is not integrated by the judicial exception because the additional elements are as follows:
“generating, by a generative adversarial network (GAN) and based on the set of training data and the fairness profile, […], a set of synthetic training data” MPEP 2106.05(f)(h) adding the words ‘apply it’ with the judicial exception, or generally linking the use of the judicial exception to a particular technological environment or field of use
“computer-implemented” MPEP 2106.05(f) merely uses a computer as a tool to perform an abstract idea, e.g. [0092] “general purpose computer”
Balance of the claim concerns GAN generating synthetic data and computer implementation. The GAN does not meaningfully limit the claim as it merely applies an off-the-shelf model as a drafting effort to monopolize the judicial exception. While the GAN generates data to include synthetic training data, the data is necessary output of applied GAN technological environment and does not detail particular GAN’s structure or how it is parameterized in a technical manner. In other words, the emphasis is on characterizing data as opposed to some specified weighting or the like. Additionally, the computer is recited at a high level of generality and is addressed at MPEP 2106.04(a)(2)(III)(c) “A claim that requires a computer may still recite a mental process.” As such, claims remain drawn to the abstract idea and additional elements fail to integrate the judicial exception into a practical application.
Step 2B: Does the claim recite additional elements that amount to significantly more than the judicial exception? No—the claims do not include additional elements that amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea in to a practical application, the additional elements are identified with respect to MPEP 2106.05 and do not reveal an inventive concept. In particular, additional elements are as follows:
“generating, by a generative adversarial network (GAN) and based on the set of training data and the fairness profile, […], a set of synthetic training data” MPEP 2106.05(f)(h) adding the words ‘apply it’ with the judicial exception, or generally linking the use of the judicial exception to a particular technological environment or field of use. Particularly, the GAN does not meaningfully limit the claim under MPEP 2106.05(e) because it merely applies a known model as a drafting effort to monopolize the judicial exception and further fails to detail a technical solution beyond that which is already established in the field of endeavor. Supplemental evidence corroborates such finding as per Tan and Rajabi, detailed below.
“computer-implemented” MPEP 2106.05(f) merely uses a computer as a tool to perform an abstract idea, e.g. [0092] “general purpose computer.” Particularly, the computer does not qualify as a particular machine under MPEP 2106.05(b).
Significantly more is not satisfied by the additional elements for at least the reasons identified, and especially when considering evidence of known FairGANs like Tan and Rajabi, detailed below. If the claim language provides only a result-oriented solution, with insufficient detail for how a computer accomplishes it, then the claims do contain an inventive concept.
Considered as a whole, looking at the limitations as an ordered combination does not elevate the claim to remedy eligibility. The skilled artisan would read claims in light of drawings Figs 5 and 4 presenting a GAN-510 with some pre-processing and applied to fairness. No strict definitions are given or otherwise required for the correlations, variables and profile upon which the claims are drafted. There is no indication that the combination of elements imparts a particular transformation or would lead to an improvement in the functioning of a computer or any other technology. Their collective functions merely provide conventional computer implementation.
In view of the foregoing, claim 1 is found to be ineligible for patent. The rejection applies to independent claims 11 and 16 as well as dependent claims 2-10, 12-15 and 16-20. Dependent claims are not found to be patent eligible because the further limitations further embellish the abstract idea and additional elements do not integrate the judicial exception into a practical application or amount to significantly more.
Independent claims 11 and 16 further recite general computer components which are treated as additional elements. Particularly, claim 11 recites “processor; and a computer-readable storage medium communicatively coupled to the processor and storing program instructions” and claim 16 recites “computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processing unit.” These additional elements fall under MPEP 2106.05(f) mere instructions to implement an abstract idea on a computer, recited at a high level of generality, and which more particularly does not qualify as a particular machine under MPEP 2106.05(b). Therefore, the additional elements do not integrate the judicial exception into a practical application or amount to significantly more.
Dependent claims 2, 12 and 17 disclose recalculating fairness score and creating fairness profile to include changes in fairness score. The limitations are considered part of the abstract idea including mathematical calculations. The repetitive calculation reflects changes in score relating to fairness which may be performed as updating of a math function. There are no additional elements.
Dependent claim 3 discloses determining Kolmogorov-Smirnov (KS) test when a difference in cumulative distribution is below a threshold. This is considered part of the abstract idea being a statistical math function as subtractive difference greater than max supremum of distributional values. KS test is formally a non-parametric statistical function published in 1958 by Herbert David1 and which has been widely used as a statistical math function. There are no additional elements.
Dependent claim 4 discloses wherein the cumulative distribution of claim 3 being below threshold indicates the two datasets are from a common set of data. This is part of the abstract idea to include math relationship such as set membership to indicate overlapping datasets may come from a joint distribution or data from the same function. There are no additional elements.
Dependent claims 5, 14 and 19 disclose wherein the correlation is based on comparing expected outcome to actual outcome for the variables. This is considered part of the abstract idea to include mathematical relationships or mental evaluation. For example, comparing may be denoted operand greater than ‘ > ‘ expectation of estimated result relative to ground truth label, or a mental comparison between hypothesis and desired target. There are no additional elements.
Dependent claim 6 discloses wherein each variable is converted to binary by gaussian mixture model, and correlation is based on comparing to include analyzing independent groups against dependent groups. This is considered part of the abstract idea being mathematical calculations. For example, binary quantization or one-hot encoded vector for model of bellman/normal distribution and evaluative comparison with analytics of sorted sets/groups. There are no additional elements.
Dependent claims 7, 13 and 18 disclose training the learning model with synthetic set of training data. The limitation is considered additional elements which amounts to generally linking the use of the judicial exception to a particular technological environment of field of use under MPEP 2106.05(h). The training to generate synthetic data is generated by generator which is the G of GAN such that any GAN would produce commensurate data, and thus does not distinguish from the field of endeavor. As such, no inventive concept is apparent and the additional elements fail to integrate the judicial exception into a practical application or amount to significantly more.
Dependent claims 8, 15 and 20 disclose wherein the GAN has generator and two discriminators. The limitation is considered additional elements which amount to adding insignificant extra-solution activity to the judicial exception under MPEP 2106.05(g). Particularly, said extra-solution activity is a well-understood, routine and conventional activity under MPEP 2106.05(d) as evidenced by Xu et al., FairGAN or FairGAN+ which shows multi-discriminator GAN architectures already established in the field of endeavor. Therefore, the additional elements do not demonstrate inventive concept and fail to integrate the judicial exception into a practical application or amount to significantly more.
Dependent claims 9-10 embellish claim 8 and disclose wherein the first discriminator identifies synthetic data, the second discriminator identifies correct outcome, and the generator generates inputs. The limitations are considered additional elements which fall under MPEP 2106.05(f) adding insignificant extra-solution activity to the judicial exception, and which are well-understood, routine and conventional activities under MPEP 2106.05(d) as evidenced by Xu at [P.1403 ¶2-3] as detailed below. No inventive concept is apparent and the additional elements fail to integrate the judicial exception into a practical application or amount to significantly more.
Considered as a whole, looking at the limitations as an ordered combination does not elevate the claim to remedy eligibility. The skilled artisan would read claims in light of the drawings Figs 5 and 4. No strict definitions are given or otherwise required for the correlations, variables and profile upon which the claims are drafted. There is no indication that the combination of elements imparts a particular transformation or would lead to an improvement in the functioning of a computer or any other technology. Their collective functions merely provide conventional computer implementation.
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.
Claims 1-2, 5-7, 11-14 and 16-19 are rejected under 35 U.S.C. 103 as being unpatentable over:
Rajabi et Garibay, “Distance Correlation GAN: Fair Tabular Data Generation with Generative Adversarial Networks” hereinafter Rajabi (TabFairGAN) in view of
Tan et al., “Improving the Fairness of Deep Generative Models without Retraining” hereinafter Tan (arXiv: 2012.04842v2).
With respect to claim 1, Rajabi teaches
A computer-implemented method {Rajabi [P.433] “Methodology” for TabFairGAN [P.435] with Algorithm [P.438] implementation by computer is [P.440 ¶] “NVIDIA GeForce RTX 3090” is a known GPU} comprising:
identifying a set of training data comprising at least one independent variable and a dependent variable {Rajabi [P.440 ¶1,3] “training set of dataset D … D = {X, S, Y}” notation introduced [P.433-34 Sect 3.1] “D = {X, S, Y} …target attribute, conditioned on the protected attribute” where the target attribute & protected attribute conveys dependent & independent variables when read in light of instant specification [0066,74] “focus attribute… protected attribute”. Rajabi describes e.g. [P.434-35] “D = {X, S, Y}, where Y ⊥ S” and defines condition of independence “variables A and B are independent, if and only if dCov2(A,B) = 0” else dependent, sets condition when independence is and is not met};
analyzing the set of training data to identify correlations between the at least one independent variable and the dependent variable {Rajabi [P.435 Sect.3] Eq.5 “distance correlation between A and B …we use distance correlation between target attribute Y and protected attribute S” again at [P.437-38] Eq.11, Alg.1 Line 12 training algorithm, analyzing is by calculation of outer for-loop for updating generator, Alg.1 Lines10-11 assign A & B variables to S & Y attributes respectively};
identifying at least one correlation between a first independent variable of the one or more independent variables and the dependent variable {Rajabi [P.438 Sect.4] “sample distance correlation” identified by sampling batched data Alg.1 Lines 9 for the distance correlation Line 12 using A/S and B/Y Lines 10-11, distance correlation detailed Eq.11 [P.438], shown Fig 1 “DistCor” introduced Eq.5 [P.435]};
calculating a fairness score for each identified correlation, including for the first independent variable, against the dependent variable {Rajabi [P.440 ¶3] “To evaluate fairness of the generated data, we use demographic parity (described in Sect. 3.2)” [P.434 Sect3.2] “quantify fairness” Eqs.1-2 scoring for a classifier of GAN, the classifier scoring regards Eq.9 critic/classifier (i.e. discriminator) of GAN subject to generator Eq.10 which controls fairness by lambda coefficient λf product with DistCor being the distance correlation, Alg.1 Line14};
creating, based on the analyzing, a fairness profile for the set of training data {Rajabi see [P.437-8 Sec4.4] “generator distribution Pg” statistical distribution is profile created by generator for training of the GAN to control fairness, see e.g. [P.435-6 Sect4.1] TabFairGAN “training by adding an extra term which calculates demographic parity, the network is fine-tuned to achieve a balance of fairness… fairness enforcement term added to the value function of generator in the second phase of training is enforcing demographic parity” see Fig 1, and Alg.1 Lines12-14};
generating, by a generative adversarial network (GAN) and based on the set of training data and the fairness profile, and in response to the first fairness score being below a fairness threshold, a set of synthetic training data {Rajabi Fig 1 “generative adversarial network” builds on TabFairGAN as author’s earlier work, Training [P.440 Sect5.3] “generated synthetic data D = {X, S, Y} …trained on the synthetic data” synthetic data for training which is generated by generator which includes profile as a statistical distribution and describes demographic parity to evaluate fairness according to classifier, similarly described at [P.435-46 Sect4.1] and detailed [P.437-38 Sect4.4]. Additionally, the threshold may comprise max of gumbel-softmax Eq.7 [P.437 Sect4.3] network architecture of TabFairGAN},
However, Rajabi does not appear to disclose the following limitation which is met by Tan:
wherein the GAN is configured to increases the fairness score for the first independent variable with above a fairness threshold {Tan discloses [P.4 Sect 3.3] “improve the fairness of GANs” uses “λa > 0 is a predefined scoring threshold” where operand ‘ > ‘ greater than is above threshold, e.g. cont’d “ai > 0 if ai = 1 otherwise ai < 0 … λai for arbitrary attribute i” attributes are introduced [P.3 ¶3] “trained GAN model… target binary attributes At …binary context attributes Ac” Eq.2 uses Ac|At which is conditional probability as a correlation. Additionally see Fig 2 GMM latent distribution for fair sampling described [P.5 Sect 3.3.2]}.
Tan is directed to Fairness with GANs thus being analogous. A person having ordinary skill in the art would have considered it obvious prior to the effective filing date to combine the teachings of Tan with Rajabi to arrive at the invention as claimed for motivation [P.2 ¶4-6] “improve the fairness of GANs not only regarding a single attribute but also across multiple attributes” includes “improving the fairness of well-learned GANs without retraining… instead of training new models or tuning the model weights, which can be often costly. In this way, our approach can be applied to numerous existing models with minor effort… Our algorithm is flexible such that we can easily include a new subgroup of interests and still maintain the fairness.”
With respect to claim 2, the combination of Rajabi and Tan teaches the computer-implemented method of claim 1, wherein the synthetic set of training data includes the at least one independent variable and the dependent variable, the method further comprising:
recalculating the fairness score for the synthetic set of training data {Rajabi [P.438] Alg.1 Line6 Update recalculates by Line7, detailed [P.437 Sect4.4] training by gradient descent, Eq.9 results in updated fairness score recalculated by updated critic network}; and
creating a synthetic data fairness profile for the synthetic set of training data, wherein the synthetic data fairness profile includes changes in the fairness score {Rajabi [P.438] Alg.1 Line13 generator update by Line 14 detailed [P.437 Sect4.4] Eq.10, the generator generates/creates synthetic data [P.440 Sect5.3] and profile is a statistical distribution subject to update}.
With respect to claim 5, the combination of Rajabi and Tan teaches the computer-implemented method of claim 1, wherein
the at least one correlation is based on comparing an expected outcome to an actual outcome for the at least one independent variable to the dependent variable {Rajabi [P.437] Eq.9 where E is expectation and comparing is difference, E subscripts r-real and g-generated are actual and expected, the generated is by generator Eq.10 where DistCor is the distance correlation}.
With respect to claim 6, the combination of Rajabi and Tan teaches the computer-implemented method of claim 5. Tan teaches wherein
each variable is converted to a binary variable by a gaussian mixture model, and the correlation is further based on a comparing that includes analyzing both independent input groups against both dependent output groups {Tan [P.3 ¶2] “binary attributes… mapped to binary labels with unit step function H(∙)” by a=H(fS(gθ(z))) where z is latent codes with Fig 2 “Gaussian Mixture Model” which “supports conditional sampling for any particular subgroup” detailed [P.5 Sect 3.2.2] “(GMM) to fit its distribution in the latent space with the set of latent code Zedit created with Eq.5 …train a GMM model on Z’edit with expectation-maximization (EM) algorithm to obtain a probabilistic model of latent codes qϕ(z) conditioned on the specific subgroup a”}.
A person having ordinary skill in the art would have considered it obvious prior to the effective filing date to specify Rajabi’s one-hot (i.e. binary) encoding [P.436 ¶3] with binarization of Tan using a Gaussian Mixture Model of Tan to arrive at the invention as claimed for a motivation [P.5 Sect 3.3.2] “To support re-sampling latent code from a specific subgroup… This model enables us to sample an arbitrary number of high-quality images from a certain subgroup” which may allow one to “filter out less confident samples… conditioned on the specified subgroup.” The latent sampling to fit GMM serves the primary goal or main objective of improving fairness with generative models including GANs [P.4 Sect 3.3], Title.
With respect to claim 7, the combination of Rajabi and Tan teaches the computer-implemented method of claim 1, further comprising:
training the learning model with the synthetic set of training data {Rajabi discloses [P.434 ¶1] “trained classifier on the synthetic data must be fair (the main objective” e.g. [P.440 ¶1,3] “synthetic data D is then used to train… classifier trained on synthetic data” similar [P.436 ¶1]}.
With respect to claim 11, the rejection of claim 1 is incorporated. The difference in scope being a system comprising processor coupled to computer-readable storage medium storing program instructions executable by processor to perform limitations of method claim 1. Rajabi discloses per [P.440 ¶4] “NVIDIA GeForce RTX 3090” is a GPU processor known specs include 24 GB DDR6 VRAM memory as a computer system environment for implementing the instructions of Algorithm 1 [P.438]. The remainder of this claim is rejected for the same rationale as claim 1.
With respect to claim 12, the combination of Rajabi and Tan teaches the system of claim 11, and further teaches the limitations of claim 2. Therefore, the rejection of claim 2 is applied to claim 12.
With respect to claim 13, the combination of Rajabi and Tan teaches the system of claim 12, and further teaches the limitations of claim 7. Therefore, the rejection of claim 7 is applied to claim 13.
With respect to claim 14, the combination of Rajabi and Tan teaches the system of claim 11, and further teaches the limitations of claim 5. Therefore, the rejection of claim 5 is applied to claim 14.
With respect to claim 16, the rejection of claim 1 is incorporated. The difference in scope being a computer program product comprising computer readable storage medium storing instructions executable by processing unit to perform limitations of method claim 1. Rajabi discloses [P.438] Algorithm instructions which are processed by [P.440 ¶4] “NVIDIA GeForce RTX 3090” GPU processor known specs include 24 GB DDR6 VRAM memory as a computer environment. The remainder of this claim is rejected for the same rationale as claim 1.
With respect to claim 17, the combination of Rajabi and Tan teaches the computer program product of claim 16, and further teaches the limitation of claim 2. Therefore, the rejection of claim 2 is applied to claim 17.
With respect to claim 18, the combination of Rajabi and Tan teaches the computer program product of claim 16, and further teaches the limitation of claim 7. Therefore, the rejection of claim 7 is applied to claim 18.
With respect to claim 19, the combination of Rajabi and Tan teaches the computer program product of claim 16, and further teaches the limitation of claim 5. Therefore, the rejection of claim 5 is applied to claim 19.
Claims 3-4 are rejected under 35 U.S.C. 103 as being unpatentable over Rajabi and Tan in view of Conway et al., US PG Pub No 2023/0325757A1 hereinafter Conway.
With respect to claim 3, the combination of Rajabi and Tan teaches the computer-implemented method of claim 2. Conway teaches further comprising:
determining, using a Kolmogorov-Smirnov (KS) test, when a difference in a cumulative distribution between the set of training data and the synthetic set of training data is below a distribution threshold {Conway [0070] “Kolmogorov-Smirnov test can be implemented to identify an upper limit (formally supremum) for the difference between two cumulative distribution functions” upper limit/supremum is the threshold, generative machine leaning models are used [0085] such that generated data is synthetic and training is by learning to provide the cumulative distribution functions}.
Conway is directed to testing generative models and considers fairness for statistical distribution thus being analogous. A person having ordinary skill in the art would have considered it obvious prior to the effective filing date to use KS-test per Conway in combination to arrive at the invention as claimed as applying known techniques to known methods ready for improvement to yield predictable results and/or for a motivation [0070] “Kolmogorov-Smirnov test provides a generalised way to compare two sample distributions… it can be used regardless of the specifics of the model being tested” again [0020].
With respect to claim 4, the combination of Rajabi, Tan and Conway teaches the computer-implemented method of claim 3, wherein
the cumulative distribution being below the distribution threshold indicates the set of training data and the synthetic set of training data are from a common set of data {Conway discloses [0070] “two cumulative distribution functions… compare two sample distributions and establish a likelihood that the two samples come from a common underlying distribution” emphasis common distribution, e.g. joint or shared distribution, threshold and data as already addressed}. Motivation for combination is applied similarly as in claim 3.
Claims 8-10, 15 and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Rajabi and Tan in view of
Xu et al., “FairGAN+: Achieving Fair Data Generation and Classification through Generative Adversarial Nets” hereinafter Xu.
With respect to claim 8, the combination of Rajabi and Tan teaches the computer-implemented method of claim 1. Xu teaches wherein
the GAN comprises a generator, a first discriminator, and a second discriminator {Xu Figs 1-2 illustrate FairGAN (2 discriminators), and FairGAN+ (3 discriminators) improved FairGAN}.
Xu is directed to Fairness with GANs thus being analogous. A person having ordinary skill in the art would have considered it obvious prior to the effective filing date to include multiple discriminators per Xu in combination to arrive at invention as claimed for motivation [P.1401 ¶1,3] “adds an additional discriminator into the original GAN to train the generator to generate fair data through adversarial learning… In order to train the classifier to be fair, we adopt another discriminator” which benefits by [P.1402 Last2¶] “By playing adversarial games with both discriminators, the generator can generate high quality fake samples” e.g. [P.1403 Last¶] “take advantage of the feedback loop of D2, G, D3, η. Improving the fairness of G can improve n on making fair predictions… FairGAN+ can perform better than a standalone generative model or standalone classifier.”
With respect to claim 9, the combination of Rajabi, Tan and Xu teaches the computer-implemented method of claim 8, wherein
the first discriminator is configured to identify synthetic training data, and the second discriminator is configured to identify a correct outcome of the dependent variable based on inputs of the one or more independent variables {Xu [P.1406 ¶3] “The first discriminator is trained to identify whether samples are real or fake. The second discriminator is trained to distinguish whether the generated samples are from the protected group or unprotected group” detailed [P.1403 ¶2-5] “discriminator D1 is trained to distinguish… discriminator D2 is trained to distinguish” distinguish/identify where synthetic data is generated by generator (to be determined real/fake by D1) and “D2 is trained to correctly predict S given a generated sample” correct outcome/prediction upon given/input samples that [P.1404 ¶1] “enforce independence between the class label Y and the protected attribute S”}. Motivation for combination is applied similarly as in claim 8.
With respect to claim 10, the combination of Rajabi, Tan and Xu teaches the computer-implemented method of claim 9, wherein
the generator is configured to generate the inputs of the one or more independent variable associated with a generated outcome {Xu Figs 1-2 show generator G arrows indicate input and output as [P.1404 Sect IV.A ¶2] “we independently generate both the label Y and the protected attribute S”}. Motivation for combination is applied similarly as in claim 8.
With respect to claim 15, the combination of Rajabi and Tan teaches the system of claim 11, and further combination with Xu teaches the limitation of claim 8. Therefore, the rejection of claim 8 with equal motivation is applied to claim 15.
With respect to claim 20, the combination of Rajabi and Tan teaches the computer program product of claim 16, and further combination with Xu teaches the limitation of claim 8. Therefore, the rejection of claim 8 with equal motivation is applied to claim 20.
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure:
Rajabi et al., “TabFairGAN: Fair Tabular Data Generation with Generative Adversarial Networks” arXiv: 2109.00666v1 largely reproduced in Distance Correlation GAN
Wu et al., “Fair Data Generation and Machine Learning Through Generative Adversarial Networks” discloses Causal-FairGAN (builds on co-author Xu’s FairGAN)
Piao et al., US PG Pub No 2025/0182029A1 see Fig 7:S10 fairness score threshold
Islam et al., US PG Pub No 2026/0111706A1 adversarial debiasing with KLD, CCA
Li et al., “FairGAN: GANs-based Fairness-aware Learning for Recommendations with Implicit Feedback” discloses FairGAN recommender
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Chase P Hinckley whose telephone number is (571)272-7935. The examiner can normally be reached M-F 9:00 - 5:00.
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, Miranda M. Huang can be reached at 571-270-7092. 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.
/CHASE P. HINCKLEY/Examiner, Art Unit 2124
1 David, Herbert “A Three-Sample Kolmogorov-Smirnov Test” 1958
Prosecution Insights