DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
This action is responsive to the Amendment filed on August 3, 2026. Claims 8-14 and 18-20 are cancelled. Claims 1-7, 15-17, and 21-24 are pending in the case. Claims 1 and 15 are the independent claims.
This action is non-final.
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 1-7, 15-17, and 21-24 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea (mental steps) without significantly more. This judicial exception is not integrated into a practical application because any additional elements amount to implementing the abstract idea on a generic computer. The claim(s) does/do not include additional elements that are sufficient to amount to significantly more than the judicial exception.
Regarding independent claims 1 and 15, and relying on the evaluation flowchart in MPEP 2106:
Step 1 (Is the claim to a process, machine, manufacture, or composition of matter?): Yes. Claim 1 is a method (process). Claim 15 is a non-transitory computer-readable media (article of manufacture).
Step 2a Prong One (Does the claim recite an abstract idea?): Yes. Claims 1 and 15 recite:
generating, from each record in a plurality of records, a respective original inference of a plurality of original inference (a mental process of determination, such as a human mentally forming an inference for each record in a plurality (i.e. at least two) of records);
measuring, based on the plurality of inferences, a fairness of the plurality of inferences (a mental process of determination, such as a human mentally determining/measuring fairness for the inferences);
selecting, after said generating, a plurality of permuted values for a feature from a plurality of original values of the feature (a mental process of determination, such as a human mentally determining/selecting permuted feature values (i.e. how current feature values should be changed or adjusted));
generating, based on said plurality of permuted values for the feature, from each record in the plurality of records, a respective permuted inference of a plurality of permuted inferences (a mental process of determination, such as a human mentally determining or forming an inference for each record, taking into consideration the adjusted or permuted feature values);
measuring, based on said plurality of original values of the feature, a fairness of said plurality of permuted inferences (a mental process of determination, such as a human mentally determining/measuring fairness for the permuted inferences);
excluding, based on a difference between the fairness of the plurality of original inferences and the fairness of the plurality of permuted inferences, the feature (a mental process of determination, such as a human mentally determining a difference in fairness between the original inferences and the permuted inferences, and determining that a feature should be excluded).
Under the broadest reasonable interpretation, these steps may be performed mentally, using mental observation and mental determination, including by a human using a physical aid such as pen and paper, including a human mentally performing observations and mentally performing mathematical calculations, and therefore correspond to the Mental Processes grouping.
Step 2a Prong Two (Does the claim recite additional elements that integrate the judicial exception into a practical application?): No. Claims 1 and 15 additionally recite:
training a machine learning model with the plurality of records excluding…the feature (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f), i.e. performing generic training with a generic machine learning model using a set of records);
(in claim 15) one or more non-transitory computer-readable media storing instructions that, when executed by one or more processors, cause (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)).
Therefore, in view of the considerations set forth in MPEP 2106.04(d), 2106.05(a)-(c) and (e)-(h), the additional elements as disclosed above alone or in combination do not integrate the judicial exception into a practical application as they are mere insignificant extra solution activity, combined with implementing the abstract idea using generic computer components.
Step 2b (Does the claim recite additional elements that amount to siqnificantly more than the judicial exception): No. Relying on the same analysis as Step 2a Prong Two (see MPEP 2106.05.I.A: Limitations that the courts have found not to be enough to qualify as “significantly more” when recited in a claim with a judicial exception include:…Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, e.g., a limitation indicating that a particular function such as creating and maintaining electronic records is performed by a computer, as discussed in Alice Corp., 573 U.S. at 225-26, 110 USPQ2d at 1984 (see MPEP 2106.05(f));…Simply appending well-understood, routine, conventional activities previously known to the industry, specified at a high level of generality, to the judicial exception...; Adding insignificant extra-solution activity to the judicial exception, as discussed in MPEP 2106.05(g);…)), claims 1 and 15 do not recite any additional elements that amount to significantly more than the abstract idea. As discussed above, Claims 1 and 15 recite:
training a machine learning model with the plurality of records excluding…the feature (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f), i.e. performing generic training with a generic machine learning model using a set of records);
(in claim 15) one or more non-transitory computer-readable media storing instructions that, when executed by one or more processors, cause (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)).
The additional elements as discussed above, in combination with the abstract idea, are not sufficient to amount to significantly more than the judicial exception as they are well, understood, routine and conventional activity as disclosed in combination with generic computer functions and components used to implement the abstract idea.
Regarding dependent claims 2 and 16:
Step 2a Prong One: incorporates the rejection of claims 1 and 15. The claims additionally recite wherein: said feature is a first feature; said plurality of permuted inferences are first permuted inferences; the method further comprises: generating, based on permuting a second feature, second permuted inferences from the plurality of records; measuring, based on said permuting the second feature, a fairness of said second permuted inferences (a mental process of determination, such as a human mentally determining additional permutations for a second feature, mentally determining additional permuted inferences based on the permutations for the second feature, and mentally determining/measuring a fairness for the additional permuted inferences).
Step 2a Prong Two: claim 16 additionally recites the instructions further cause (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)).
Step 2b: claim 16 additionally recites the instructions further cause (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)).
Regarding dependent claims 3 and 21:
Step 2a Prong One: incorporates the rejection of claims 2 and 16. The claims additionally recite excluding, based on a difference between a fairness of the plurality of original inferences and the fairness of the second permuted inferences, the second feature (a mental process of determination, such as a human mentally determining a difference in fairness between the original inferences and the second permuted inferences, and determining that a feature should be excluded).
Step 2a Prong Two: the claims additionally recite training the machine learning model excluding…the second feature and (in claim 21) the instructions further cause (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)).
Step 2b: the claims additionally recite training the machine learning model excluding…the second feature and (in claim 21) the instructions further cause (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)).
Regarding dependent claims 4 and 22:
Step 2a Prong One: incorporates the rejection of claims 3 and 21.
Step 2a Prong Two: the claims additionally recite wherein said training excluding the first feature and said training excluding the second feature are a same training (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)).
Step 2b: the claims additionally recite wherein said training excluding the first feature and said training excluding the second feature are a same training (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)).
Regarding dependent claims 5 and 23:
Step 2a Prong One: incorporates the rejection of claims 3 and 21; the claims further recite wherein: said permuting the second feature is a first permuting the second feature; the method further comprises before said training excluding the first feature, generating third permuted inferences from the plurality of records based on a second permuting the second feature…includes, based on a fairness of said third permuted inferences, the second feature (a mental process of determination, such as a human mentally determining permutation/adjustment to a second feature a first time and a second time, mentally determining third permuted inferences based on the determined permutation/adjustment to the second feature form the second time, and then determining whether or not to include the second feature based on a determined fairness of the third permuted inferences).
Step 2a Prong Two: the claims additionally recite said training excluding the first feature includes…the second feature; said training excluding the second feature excludes the first feature and (in claim 23) the instructions further cause (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)).
Step 2b: the claims additionally recite said training excluding the first feature includes…the second feature; said training excluding the second feature excludes the first feature and (in claim 23) the instructions further cause (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)).
Regarding dependent claims 6 and 17:
Step 2a Prong One: incorporates the rejection of claims 1 and 15; the claims further recite measuring a fairness of inferences from the plurality of records based on said training excluding the feature; said fairness of inferences based on said training excluding the feature is higher than said fairness of said plurality of original inferences (a mental process of evaluation, such as a human mentally determining or measuring a fairness of inferences based on/after training excluding the feature (such as mentally observing outputs of a corresponding trained model and mentally determining a corresponding fairness of the outputs), where the resulting fairness is judged to be higher than the original fairness).
Step 2a Prong Two: the claims additionally recite said generating said plurality of original inferences comprises training the machine learning model including the feature and (in claim 17) the instructions further cause (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f) and a field of use and technological environment as discussed in MPEP 2106.05(h)).
Step 2b: the claims additionally recite said generating said plurality of original inferences comprises training the machine learning model including the feature and (in claim 17) the instructions further cause (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f) and a field of use and technological environment as discussed in MPEP 2106.05(h)).
Regarding dependent claims 7 and 24:
Step 2a Prong One: incorporates the rejection of claims 1 and 15.
Step 2a Prong Two: the claims additionally recite wherein the machine learning model does not comprise a random forest or a decision tree (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f) and a field of use and technological environment as discussed in MPEP 2106.05(h)).
Step 2b: the claims additionally recite wherein the machine learning model does not comprise a random forest or a decision tree (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f) and a field of use and technological environment as discussed in MPEP 2106.05(h)).
Therefore, in view of the considerations set forth in MPEP 2106.04(d), 2106.05(a)-(c) and (e)-(h), the additional elements as recited in the dependent claims discussed above alone or in combination do not integrate the judicial exception into a practical application as they are mere insignificant extra solution activity, combined with implementing the abstract idea using generic computer components, and limitations describing a field of use or technological environment. The additional elements as discussed above, in combination with the abstract idea, are not sufficient to amount to significantly more than the judicial exception as they are well, understood, routine and conventional activity as disclosed in combination with generic computer functions and components used to implement the abstract idea, and limitations describing a field of use or technological environment.
Claim Rejections – 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries set forth in Graham v. John Deere Co., 383 U.S. 1, 148 USPQ 459 (1966), that are applied 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 under pre-AIA 35 U.S.C. 103(a), the examiner presumes that the subject matter of the various claims was commonly owned at the time any inventions covered therein were made absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and invention dates of each claim that was not commonly owned at the time a later invention was made in order for the examiner to consider the applicability of pre-AIA 35 U.S.C. 103(c) and potential pre-AIA 35 U.S.C. 102€, (f) or (g) prior art under pre-AIA 35 U.S.C. 103(a).
Claims 1-7, 15-17, and 21-24 are rejected under 35 U.S.C. 103 as being unpatentable over Yao et al. (US 20250013887 A1) in view of Hesami et al. (US 20220027986 A1).
With respect to claims 1 and 15, Yao teaches one or more non-transitory computer-readable media storing instructions that, when executed by one or more processors, cause performance of a method; and the method, comprising:
generating, from each record in a plurality of records, a respective original inference of a plurality of original inferences (e.g. paragraph 0024, original model based on original sample set and trained using a loss function; paragraph 0028, Figs. 1 and 2, calculating fairness of original model on validation sample set; performance of original model on the real data to examine performance of the model in the real environment; i.e. the validation sample set is applied to the original model (that is, the original model performs prediction/inference on the validation sample set) in order to determine the corresponding fairness metric; Examiner additionally notes that at least claims 6 and 17 of the instant application appear to indicate that the recited act of generating an inference can include training of the model);
measuring, based on the plurality of original inferences, a fairness of the plurality of original inferences (e.g. paragraph 0026, determining fairness metric 132 of the original model on the validation sample set 126; paragraph 0028, Figs. 1 and 2, determining fairness metric 132 of original model on validation sample set; fairness metric quantifying fairness of original model on validation sample set);
selecting, after said generating, a plurality of permuted values for a feature from a plurality of original values of the feature (e.g. paragraph 0025, choosing feature of sample set and counterfactually intervening on the feature to answer the question “how will the fairness metric of the model change if values of some features of the training samples of the model are changed”; generating adjusted sample set comprising counterfactual samples and original samples; paragraph 0027, Figs. 1 and 2, generating counterfactual sample by adjusting original samples in original sample set, the original sample set used to generate the original model for performing classification task; original sample set used to generate the original model; intervening on the concepts (including features) of the original samples in the original sample set to thereby generate the counterfactual sample; paragraph 0034, Fig. 3C, discussing intervening on feature of sample; intervention on the feature means that the model will become fairer when the set of collected training sample has different values for the same feature; paragraph 0058, Fig. 5B, intervening on feature of original sample, changing value of feature; generating new features and labels for counterfactual sample);
generating, based on said plurality of permuted values for the feature, from each record in the plurality of records, a respective permuted inference of a plurality of permuted inferences (e.g. paragraph 0026, counterfactual model is model corresponding to adjusted sample set, and may be counterfactual model obtained if adjusted sample set is used for training; paragraph 0029, Figs. 1-2, fairness metric 134 of the counterfactual model on the validation sample set; i.e. the validation sample set is applied to the counterfactual model (that is, the counterfactual model performs prediction/inference on the validation sample set) in order to determine the corresponding fairness metric; Examiner additionally notes that at least claims 6 and 17 of the instant application appear to indicate that the recited act of generating an inference can include training of the model);
measuring, based on said plurality of original values of the feature, a fairness of said plurality of permuted inferences (e.g. paragraph 0026, determining fairness metric 134 of the counterfactual model on the validation sample set 126; paragraph 0029, fairness metric 134 of the counterfactual model on the validation sample set); and
training a machine learning model with the plurality of records excluding, based on a difference between the fairness of the plurality of original inferences and the fairness of the plurality of permuted inferences, the feature (e.g. paragraph 0026, determining fairness impact value 136 based on the fairness metrics 132 and 134; paragraph 0029, determining fairness impact of original sample on original model based on the fairness metric, the original sample, and the counterfactual sample; fairness impact value 136 indicating difference between the fairness metric 132 of the original model and the fairness metric 134 of the counterfactual model; paragraph 0065, target samples having greatest fairness impact values determined; updated sample set generated by removing plurality of target samples from original sample set and updated model generated based on the updated sample set; where one or more features have been permuted, and the corresponding sample or samples are removed from the training data, this includes excluding at least the features which are included in the removed samples/to the extent that they are removed along with the corresponding sample).
Assuming arguendo that Yao does not explicitly disclose the plurality of records excluding the feature (i.e. in that Yao teaches removing the sample, which includes the permuted/altered feature, from the sample set but does not explicitly disclose removing the feature itself), Hesami teaches the plurality of records excluding the feature (e.g. paragraphs 0090-0091, removing features from the training data).
Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the invention having the teachings of Yao and Hesami in front of him to have modified the teachings of Yao (directed to determining fairness impact of a model), to incorporate the teachings of Hesami (directed to model training, including based on bias/fairness) to include the capability to remove the feature from the sample set/plurality of records. One of ordinary skill would have been motivated to perform such a modification in order to improve the fairness of the model as described in Hesami (paragraph 0091).
With respect to claims 2 and 16, Yao in view of Hesami teaches all of the limitations of claims 1 and 15 as previously discussed, and Yao further teaches wherein:
said feature is a first feature (e.g. paragraph 0025, choosing feature of sample set and counterfactually intervening on the feature to answer the question “how will the fairness metric of the model change if values of some features of the training samples of the model are changed”; generating adjusted sample set comprising counterfactual samples and original samples; paragraph 0027, Figs. 1 and 2, generating counterfactual sample by adjusting original samples in original sample set, the original sample set used to generate the original model for performing classification task; original sample set used to generate the original model; intervening on the concepts (including features) of the original samples in the original sample set to thereby generate the counterfactual sample; paragraph 0034, Fig. 3C, discussing intervening on feature of sample; intervention on the feature means that the model will become fairer when the set of collected training sample has different values for the same feature; paragraph 0058, Fig. 5B, intervening on feature of original sample, changing value of feature; generating new features and labels for counterfactual sample);
said plurality of permuted inferences are first permuted inferences (e.g. paragraph 0026, counterfactual model is model corresponding to adjusted sample set, and may be counterfactual model obtained if adjusted sample set is used for training; paragraph 0029, Figs. 1-2, fairness metric 134 of the counterfactual model on the validation sample set; i.e. the validation sample set is applied to the counterfactual model (that is, the counterfactual model performs prediction/inference on the validation sample set) in order to determine the corresponding fairness metric; Examiner additionally notes that at least claims 6 and 17 of the instant application appear to indicate that the recited act of generating an inference can include training of the model);
the method further comprises:
generating, based on permuting a second feature, second permuted inferences from the plurality of records (e.g. paragraph 0026, counterfactual model is model corresponding to adjusted sample set, and may be counterfactual model obtained if adjusted sample set is used for training; paragraph 0029, Figs. 1-2, fairness metric 134 of the counterfactual model on the validation sample set; paragraph 0058, Fig. 5B, generating counterfactual sample by intervening in feature of original sample; intervening in feature 505, such as by changing its value, to produce feature 525; when one feature of sample is changed, other features are affected at the same time; generating new features and labels for the counterfactual sample, inputting feature 525, 506, and 507 into GAN which outputs new feature 526 and 527; inputting features 525, 526, and 527 into original model to output new label and obtain counterfactual sample; paragraph 0065, respectively changing values of specified features of plurality of original samples; i.e. multiple features may be permuted/changed, including at least a second feature, and training/inferencing using these additional permuted/changed features may also be performed);
measuring, based on said permuting the second feature, a fairness of said second permuted inferences (e.g. paragraph 0026, determining fairness metric 134 of the counterfactual model on the validation sample set 126; paragraph 0029, fairness metric 134 of the counterfactual model on the validation sample set; paragraph 0065, determining plurality of target samples have greatest corresponding fairness impact; i.e. where multiple features have been permuted/changed, a fairness metric may be determined taking these into account as well).
With respect to claims 3 and 21, Yao in view of Hesami teaches all of the limitations of claims 2 and 16 as previously discussed, and Yao further teaches the method further comprising training the machine learning model excluding, based on a difference between the fairness of the plurality of original inferences and the fairness of the second permuted inferences, the second feature (e.g. paragraph 0026, determining fairness impact value 136 based on the fairness metrics 132 and 134; paragraph 0029, determining fairness impact of original sample on original model based on the fairness metric, the original sample, and the counterfactual sample; fairness impact value 136 indicating difference between the fairness metric 132 of the original model and the fairness metric 134 of the counterfactual model; paragraph 0058 and Fig. 5B, indicating that multiple features may be permuted along with a first feature which is selected for permutation; paragraph 0065, updated sample set generated by removing plurality of target samples from the original sample set, and an updated model may be generated based on the updated sample set; i.e. where multiple features have been permuted, and the corresponding sample or samples are removed from the training data, this includes excluding at least the multiple features which are included in the removed samples/to the extent that they are removed along with the corresponding sample).
With respect to claims 4 and 22, Yao in view of Hesami teaches all of the limitations of claims 3 and 21 as previously discussed, and Yao further teaches wherein said training excluding the first feature and said training excluding the second feature are a same training (e.g. paragraph 0058 and Fig. 5B, indicating that multiple features may be permuted along with a first feature which is selected for permutation; paragraph 0065, updated sample set generated by removing plurality of target samples from the original sample set, and an updated model may be generated based on the updated sample set; i.e. where a plurality of samples/features are removed from the dataset and this is used for retraining/updating the model, this includes training excluding both first and second features in a same training; i.e. where multiple features are permuted within a set of samples, and those samples including the multiple features are removed/excluded from the training data used for subsequent training, this includes excluding the multiple features from the same training).
With respect to claims 5 and 23, Yao in view of Hesami teaches all of the limitations of claims 3 and 21 as previously discussed, and Yao further teaches wherein:
said permuting the second feature is a first permuting the second feature (e.g. paragraph 0058, Fig. 5B, generating counterfactual sample by intervening in feature of original sample; intervening in feature 505, such as by changing its value, to produce feature 525; when one feature of sample is changed, other features are affected at the same time; generating new features and labels for the counterfactual sample, inputting feature 525, 506, and 507 into GAN which outputs new feature 526 and 527; inputting features 525, 526, and 527 into original model to output new label and obtain counterfactual sample; paragraph 0065, respectively changing values of specified features of plurality of original samples; i.e. multiple features may be permuted/changed, including at least a second feature);
the method further comprises before said training excluding the first feature, generating third permuted inferences from the plurality of records based on a second permuting the second feature (e.g. paragraph 0026, counterfactual model is model corresponding to adjusted sample set, and may be counterfactual model obtained if adjusted sample set is used for training; paragraph 0029, Figs. 1-2, fairness metric 134 of the counterfactual model on the validation sample set; paragraph 0030, making adjustments or intervention to each original sample in the original sample set; paragraph 0049, plurality of counterfactual samples are intervened at a time; paragraph 0054, Fig. 5A, process intervening on population attribute 504 of original sample; when population attribute is changed, features and label of the sample are affected; generating new features for the counterfactual sample; inputting features 505, 506, and 507 into GAN which outputs new features 515, 516, and 517; i.e. in addition to intervening/permuting a specific selected feature (and at least one second feature which has a dependency upon that feature), a population attribute can also be intervened/permuted as part of the overall intervening of each sample in the sample set, and the corresponding features may be intervened/permuted based on this as well (i.e. resulting in at least a second permuting/intervening on at least one second feature), and all of these features may be utilized as part of the adjusted sample set for training the counterfactual model (i.e. where the training of the counterfactual model includes the model performing inferencing using the permuted examples in the adjusted sample set, resulting in respective first, second, and third permuted inferences corresponding to each of the first, second, and third permuted features));
said training excluding the first feature includes, based on a fairness of said third permuted inferences, the second feature (e.g. paragraph 0065, determining plurality of fairness impact values corresponding to plurality of original samples by respectively changing values of specified features, etc.; the plurality of target samples having the greatest corresponding fairness impact values determined from the plurality of original samples based on the plurality of fairness impact values, and updated sample set is generated by removing the plurality of target samples from the original sample set and updated model is generated based on the updated sample set; i.e. based on fairness values corresponding to each sample having different sets of permuted/intervened feature values, a first sample corresponding to a first permuted feature may be determined as a target sample based on having a greatest fairness impact value and may therefore be excluded from the updated sample set, while a second sample corresponding to a second/third permuted feature may be determined as not being a target sample based on not having a sufficiently high fairness impact value and may therefore be included in the updated sample set, and the updated sample set (excluding the first feature while including the second feature) may then be used to train/generate the updated model);
said training excluding the second feature excludes the first feature (e.g. paragraph 0065, determining plurality of fairness impact values corresponding to plurality of original samples by respectively changing values of specified features, etc.; the plurality of target samples having the greatest corresponding fairness impact values determined from the plurality of original samples based on the plurality of fairness impact values, and updated sample set is generated by removing the plurality of target samples from the original sample set and updated model is generated based on the updated sample set; i.e. based on fairness values corresponding to each sample having different sets of permuted/intervened feature values, a first sample corresponding to a first permuted feature may be determined as a target sample based on having a greatest fairness impact value and may therefore be excluded from the updated sample set, while a second sample corresponding to a second/third permuted feature may also be determined to be a target sample based on having a sufficiently high fairness impact value and may therefore also be excluded from the updated sample set, and the updated sample set (excluding the first and second features) may then be used to train/generate the updated model).
With respect to claims 6 and 17, Yao in view of Hesami teaches all of the limitations of claims 1 and 15 as previously discussed, and Yao further teaches wherein: said generating said plurality of original inferences comprises training the machine learning model including the feature (e.g. paragraph 0024, original model based on original sample set and trained using a loss function; each original sample in the original sample set has a feature 116, etc.; i.e. the original model is trained (where this training includes generating inferences) using the original sample set, which includes the features, as shown in Fig. 1; as noted in paragraph 0109 of the specification of the instant application, the process or training the model includes the model/algorithm applying the model to a training input to generate a predicted output (i.e. an inference); therefore the process of training the original model on the original sample set includes the original model generating inferences from each sample/record in the sample set).
Yao does not explicitly disclose:
the method further comprises measuring a fairness of inferences from the plurality of records based on said training excluding the feature;
said fairness of inferences based on said training excluding the feature is higher than said fairness of said plurality of original inferences.
However, Hesami teaches:
the method further comprises measuring a fairness of inferences from the plurality of records based on said training excluding the feature (e.g. paragraphs 0090-0091, inferring labels introduces bias to the model; model trained using fairness evaluation system; to reduce bias, features removed from the training data and the model can be retrained until the effects of model biases are reduced; comparing biases against fairness criteria; if model does not satisfy fairness criteria, features removed and the model is retrained and evaluated for fairness, until fairness criteria has been satisfied; i.e. upon removing/excluding features from the training data, the model may be retrained with the updated training data, and the corresponding model is reevaluated by measuring fairness; as discussed above, the process of training/retraining the model would include the model making inferences on the training data set (modified to exclude/remove features), and measuring the fairness of the model retrained on this updated data set is therefore analogous to measuring fairness of inference from the records based on training excluding the feature);
said fairness of inferences based on said training excluding the feature is higher than said fairness of said plurality of original inferences (e.g. paragraph 0090, model is trained (using dataset with removed features) until effects of biases are reduced; paragraph 0091, training of the model to improve fairness; i.e. after training the model’s bias is reduced/fairness is improved/higher than the original model based on the original inferences).
Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the invention having the teachings of Yao and Hesami in front of him to have modified the teachings of Yao (directed to determining fairness impact of a model), to incorporate the teachings of Hesami (directed to model training, including based on bias/fairness) to include the capability to, after removing the features from the training dataset, retrain the model and measure the fairness of the retrained model. One of ordinary skill would have been motivated to perform such a modification in order to improve the fairness of the model as described in Hesami (paragraph 0091).
With respect to claims 7 and 24, Yao in view of Hesami teaches all of the limitations of claims 1 and 15 as previously discussed, Yao and Hesami further teach wherein the machine learning model does not comprise a random forest or a decision tree (e.g. Yao paragraph 0024, indicating that the original model 110 is a classification model trained using a loss function (but not indicating that it comprises a random forest or decision tree); Hesami paragraph 0070-0072, indicating that a labeling model can be a supervised model such as a gradient boosted tree while in at least one implementation, a model can be a semi-supervised model such as a self-training model, a graph-based model, or non-graph based model, and can be a KGB model, use trained autoencoders, etc.; paragraph 0090, indicating that the model trained as S240 is different from any of the labeling models trained at S230; i.e. where labeling models may be implemented as tree or forest based models, another model, which is evaluated for fairness, may be a different type of model, such as a graph-based model, a model comprising autoencoders, etc. (i.e. not a random forest or decision tree model)).
It is noted that any citation to specific pages, columns, lines, or figures in the prior art references and any interpretation of the references should not be considered to be limiting in any way. “The use of patents as references is not limited to what the patentees describe as their own inventions or to the problems with which they are concerned. They are part of the literature of the art, relevant for all they contain,” In re Heck, 699 F.2d 1331, 1332-33, 216 USPQ 1038, 1039 (Fed. Cir. 1983) (quoting in re Lemelson, 397 F.2d 1006, 1009, 158 USPQ 275, 277 (GCPA 1968)). Further, a reference may be relied upon for all that it would have reasonably suggested to one having ordinary skill the art, including nonpreferred embodiments. Merck & Co, v. Biocraft Laboratories, 874 F.2d 804, 10 USPQ2d 1843 (Fed. Cir.), cert, denied, 493 U.S. 975 (1989). See also Upsher-Smith Labs. v. Pamlab, LLC, 412 F,3d 1319, 1323, 75 USPQ2d 1213, 1215 (Fed. Cir, 2005): Celeritas Technologies Ltd. v. Rockwell International Corp., 150 F.3d 1354, 1361, 47 USPQ2d 1516, 1522-23 (Fed. Cir. 1998).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant’s disclosure.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to JEREMY L STANLEY whose telephone number is (469)295-9105. The examiner can normally be reached on Monday-Friday from 9:00 AM to 5:00 PM CST.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Abdullah Al Kawsar, can be reached at telephone number (571) 270-3169. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of an application may be obtained from Patent Center and the Private Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from Patent Center or Private PAIR. Status information for unpublished applications is available through Patent Center and Private PAIR for authorized users only. Should you have questions about access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free).
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) Form at https://www.uspto.gov/patents/uspto-automated- interview-request-air-form.
/JEREMY L STANLEY/
Primary Examiner, Art Unit 2127
Prosecution Insights