DETAILED ACTION
This Action is responsive to Claims filed 07/07/2026.
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 .
Status of the Claims
Claims 1-3, 7, 9-11, 15, and 17-18 have been amended. Claims 5-6, 13-14, and 20 have been cancelled. Claim 21 is new. Claims 1-3, 7, 9-11, 15, 17-18, and 21 are currently pending.
Response to Arguments
Applicant's arguments, see Pages 11-19, filed 07/07/2026, regarding the 35 U.S.C. 101 Rejection of Claims 1-3, 7, 9-11, 15, 17-18 have been fully considered but they are not persuasive.
The amendments to the claims are not drafted in such a way as to preclude a human mind with or without the aid of pen and paper from performing the claimed steps.
The “modifying…”, “prospecting…”, “identifying…” (and subsequent limitations refining the features), and “ranking…” steps do not recite specific structure or implementation precluding a human mind from performing the generic actions on generic “features.”
The “generating…” step amounts merely to the generic operation of a GAN, and amounts to instructions to apply the interpretable abstract idea steps resulting in the ranking.
The “altering…” step is not recited in such a way as to preclude a human mind with the aid of pen and paper from performing the claimed step. There is no recitation of specific structure or implementation beyond the generic “computer-implemented method” for altering the media. A human mind is capable of generically altering and/or obfuscating regions of generic media, by redaction, for example.
The “determining…”, “determining…”, and “selecting…” steps do not recite specific structure or implementation precluding a human mind from performing the generic actions on generic “features.”
The “performing…” step amounts to instructions to apply the aforementioned abstract idea mental process steps.
Given the above interpretation, the only limitations representing additional elements amount to instructions to apply interpretable abstract idea mental process steps that do not recite specific structure or implementation precluding a human mind from performing them. The alleged improvement the Applicant cites does not come from the generic operation of a machine learning model, but the series of algorithmic data observation/manipulation abstract idea mental process steps. Per MPEP 2106.05(a), the specific improvement cannot come from the abstract idea(s). See the updated 101 Rejection below.
Applicant’s arguments with respect to claim(s), see Pages 19-21, filed 07/07/2026, regarding the 35 U.S.C. 103 Rejection(s) of Claims 1-3, 7, 9-11, 15, 17-18 have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument.
Claim Rejections - 35 USC § 101
The text of those sections of Title 35, U.S. Code not included in this action can be found in a prior Office action.
Claims 1-3, 7, 9-11, 15, 17-18, and 21 rejected under 35 U.S.C. 101 because the claimed invention is directed to a judicial exception (i.e., a law of nature, a natural phenomenon, or an abstract idea) without significantly more; and because the claims as a whole, considering all claim elements both individually and in combination, do not amount to significantly more than the abstract idea, see Alice Corporation Pty. Ltd. v. CLS Bank International, et al, 573 U.S. (2014). In determining whether the claims are subject matter eligible, the Examiner applies the 2019 USPTO Patent Eligibility Guidelines. (2019 Revised Patent Subject Matter Eligibility Guidance, 84 Fed. Reg. 50, Jan. 7, 2019.)
Step 1:
Claims 1-3, 7, and 21 recite a computer-implemented method comprising: identifying protected features from media that enable bias in a decision-making process of an artificial intelligence (Al) system, which falls under the statutory category of a process. Claims 9-11 and 15 recite an apparatus comprising: a memory; and at least one processor, coupled to said memory, and operative to perform operations, which falls under the statutory category of a machine. Claims 17-18 recites a computer program product, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, which falls under the statutory category of a manufacture.
Step 2A – Prong 1:
Claim 1 recites an abstract idea, law of nature, or natural phenomenon. The limitations of “modifying a set of candidate features of a plurality of candidate features of a model of an artificial intelligence (Al) system using principled modification corresponding to modification-triggered prediction drift evaluation, wherein the principled modification is a modification of the set of candidate features that are known or suspected of being protected features;”, “prospecting the set of candidate features of the model based on the modifying;”, “identifying the protected features from media that enable bias in a decision-making process of the Al system,”, “the identifying comprises selecting a portion of the set of candidate features as the protected features, the selected portion of the set of candidate features has a highest divergence, during the modification-triggered prediction drift evaluation, among the plurality of candidate features, the protected features comprise biasing cues and spurious correlations, and the biasing cues and the spurious correlations are associated with a background of the media;”, “ranking the identified protected features;”, “altering the background of the media by integrating the adversarial artifacts into the background of the media and by excluding the ranked protected features from the media;”, “determining that a Kullback-Leibler divergence, used to evaluate detectability of the ranked protected features, is not less than a threshold;”, “determining, based on the determining that the Kullback-Leibler divergence is not less than the threshold, that the generative adversarial network is not able to detect each of the ranked protected features in the media after integrating the adversarial artifacts into the media;”, and “selecting, based on determining that the generative adversarial network is not able to detect each of the ranked protected features, a dynamic option or a static option for incorporating adversarial data into the media;” under the broadest reasonable interpretation, covers a mental process including an observation, evaluation, judgment or opinion that could be performed in the human mind or with the aid of pencil and paper. This limitation therefore falls within the mental process group.
Modifying a set of generic features of a generic AI model is practically performed within the human mind or with the aid of pen and paper. Prospecting, identifying, and selecting features of generic media is practically performed within the human mind or with the aid of pen and paper. Ranking features is practically performed within the human mind or with the aid of pen and paper. Generating obfuscating artifacts if generic media is practically performed within the human mind or with the aid of pen and paper. Integrating the artifacts into generic media is practically performed within the human mind or with the aid of pen and paper. Determining a Kullback-Leibler divergence and determining it is less than a threshold is practically performed within the human mind or with the aid of pen and paper. Selecting a dynamic or static option is practically performed within the human mind or with the aid of pen and paper.
Step 2A – Prong 2:
The additional elements of claim 1 do not integrate the abstract idea into a judicial exception. Claim 1 recites the additional elements “A computer-implemented method” and “media” are recognized as generic computer components recited at a high level of generality. Although it has and executes instructions to perform the abstract idea itself, this also does not serve to integrate the abstract idea into a practical application as it merely amounts to instructions to "apply it." (See MPEP 2106.04(d)(2) indicating mere instructions to apply an abstract idea does not amount to integrating the abstract idea into a practical application).
The additional elements recited in the limitations “protected features“, “a decision-making process”, “an artificial intelligence (Al) system”, “a generative adversarial network”, “adversarial artifacts” are recognized as non-generic computer components, however, they are found to generally link the abstract idea to a particular technological environment or field of use (See MPEP 2106.05(h)).
The additional element recited in the limitation “generating, using a generative adversarial network, adversarial artifacts to obfuscate at least a portion of the ranked protected features;“ and “performing, by AI system, inferencing using the media.” amounts to mere instructions to apply the abstract idea mental process steps involved in the pre-processing of the generic media (See MPEP 2106.05(f)).
Step 2B:
The only limitation on the performance of the described method is a limitation reciting “A computer-implemented method” and “media” These elements are insufficient to transform a judicial exception to a patentable invention because the recited elements are considered insignificant extra-solution activity (generic computer system, processing resources, links the judicial exception to a particular, respective, technological environment). The claim thus recites computing components only at a high-level of generality such that it amounts to no more than mere instructions to apply the exception using generic computer components; mere instructions to apply an exception using a generic computer component cannot provide an inventive concept (see MPEP 2106.05(f)).
The additional elements recited in the limitations “protected features“, “a decision-making process”, “an artificial intelligence (Al) system”, “a generative adversarial network”, “adversarial artifacts” are recognized as non-generic computer components, however, they are found to generally link the abstract idea to a particular technological environment or field of use (See MPEP 2106.05(h)).
The additional element recited in the limitation “generating, using a generative adversarial network, adversarial artifacts to obfuscate at least a portion of the ranked protected features;“ and “performing, by AI system, inferencing using the media.” amounts to mere instructions to apply the abstract idea mental process steps involved in the pre-processing of the media (See MPEP 2106.05(f)).
Taken alone or in ordered combination, these additional elements do not amount to
significantly more than the above-identified abstract idea. There is no indication that the
combination of elements improves the functioning of a computer or improves any other
technology. Their collective functions merely provide conventional computer implementation.
For the reasons above, claim 1 is rejected as being directed to non-patentable subject matter under §101. This rejection applies equally to independent claims 9 and 17.
Claim 9 recites “An apparatus comprising: a memory; and at least one processor, coupled to said memory, and operative to perform operations comprising:” (generic computer components). The additional elements found in claim 1 and repeated in claim 9 are recognized as non-generic computer components, however, they are found to generally link the abstract idea to a particular technological environment or field of use (See MPEP 2106.05(h)).
Claim 17 recites “A computer program product, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a computer to cause the computer to perform operations comprising:” (generic computer components). The additional elements found in claim 1 and repeated in claim 17 are recognized as non-generic computer components, however, they are found to generally link the abstract idea to a particular technological environment or field of use (See MPEP 2106.05(h)).
Dependent Claims:
Claim 2 (claim 10) recites additional elements “protected attributes, protected cues, and spurious correlations.” These are recognized as non-generic computer components, however, they are found to generally link the abstract idea to a particular technological environment or field of use (See MPEP 2106.05(h)).
Claim 3 (claims 11 and 18) recites abstract idea mental process steps “obtaining knowledge information on the AI system and a plurality of features from the media;” and “and determining a type characteristic of the model of the AI system, wherein the identification is based on the type characteristic of the model.”
Claim 7 (claim 15) recites additional elements which are recognized as non-generic computer components, however, are found to generally link the abstract idea to a particular technological environment or field of use (See MPEP 2106.05(h)).
Claim 21 recites refinements to the data of claim 1.
Claim Rejections - 35 USC § 103
The text of those sections of Title 35, U.S. Code not included in this action can be found in a prior Office action.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claim(s) 1-3, 7, 9-11, 15, 17-18, and 21 is/are rejected under 35 U.S.C. 103 as being unpatentable over Dalli et al. (US 2022/0012591 A1), hereinafter Dalli; Shanbhag et al. (Unified Shapley Framework to Explain Prediction Drift, 2021), hereinafter Shanbhag; Tieu et al. (Spatio-temporal generative adversarial network for gait anonymization, 2019), hereinafter Tieu; and Luo et al. (Zero Kullback-Liebler Divergence Image Data Hiding, 2011), hereinafter Luo.
In regards to claim 1: The present invention claims: “A computer-implemented method, comprising: modifying a set of candidate features of a plurality of candidate features of a model of an artificial intelligence (Al) system…” Dalli teaches “Methods for detecting bias, strength, and weakness of data sets and the resulting models may be described. A method may implement global bias detection which utilizes the coefficients of the model to identify, minimize, and/or correct potential bias within a desired error tolerance. Another method makes use of local feature importance extracted from the rule-based model coefficients to locally identify bias. A third method aggregates the feature importance over the results/explanations of multiple samples.” (Abstract).
“performing, by the AI system, inferencing using the media.” Dalli teaches the system ultimately performs inference with the updated data or model ([0079])
Dalli fails to explicitly teach “using principled modification corresponding to modification-triggered prediction drift evaluation, wherein the principled modification is a modification of the set of candidate features that are known or suspected of being protected features;” However, Shanbhag teaches “A systematic method is thus needed for studying prediction drift and attributing it to a) the features of the model and b) the individual data points that constitute the distributional samples that are compared.” (introduction) and goes on to describe an application of their algorithm in Section 9.2 (Page 7), where prediction drift is attributed to predetermined, important features.
Shanbhag highlights the importance of determining how important features of an input can effect the prediction drift of a model, and the need to ascertain which features contribute to the drift (Introduction). It would have been obvious to one of ordinary skill in the art at the time of the applicant’s filing to combine the prediction drift attribution methods of Shanbahg in a bias detection system such as Dalli’s in order to determine important features for the drift of the model.
“prospecting the set of candidate features of a model based on the modifying;” Dalli teaches “The black-box system may be a machine learning algorithm of any underlying architecture. In an exemplary embodiment, the machine learning algorithm may be a deep neural network (DNN). The black-box system may additionally contain non-linear modelled data. The underlying architecture and structure of the black box algorithm may not be important since it does not need to be analyzed directly. Instead, the training data may be loaded as input 204, and the output can be recorded as data point predictions or classifications 206. Since a large amount of broad data is loaded as input, the output data point predictions or classifications may provide a global view of the black box algorithm.
Still referring to exemplary FIG. 2, the method may continue by aggregating the data point predictions or classifications into hierarchal partitions 208. Rule conditions may be obtained from the hierarchal partitions. An external function defined by Partition(X) may identify the partitions. Partition(X) may be a function configured to partition similar data and may be used to create rules. The partitioning function may consist of a clustering algorithm such as k-means, entropy or a mutual information (MI) based method.” ([0084]-[0085], mapping the use of a black-box prediction results, which are used in the formulation of a white-box model later ([0086]), to determining which features or rules affect the bias of the overall system). Shanbahg’s method also highlights important features based on the prediction drift (Section 9.2).
“identifying the protected features from media that enable bias in a decision-making process of the AI system, wherein the identifying comprises selecting a portion of the set of candidate features as the protected features, the selected portion of the set of candidate features has a highest divergence, during the modification-triggered prediction drift evaluation, among the plurality of candidate features;” Dalli teaches “Methods for detecting bias, strength, and weakness of data sets and the resulting models may be described. A method may implement global bias detection which utilizes the coefficients of the model to identify, minimize, and/or correct potential bias within a desired error tolerance. Another method makes use of local feature importance extracted from the rule-based model coefficients to locally identify bias. A third method aggregates the feature importance over the results/explanations of multiple samples.” (Abstract). Shanbahg’s method also highlights important features based on the prediction drift (Section 9.2).
“the protected features comprise biasing cues and spurious correlations, and the biasing cues and the spurious correlations are associated with a background of the media;” Dalli teaches protected attributes ([0044]) and an embodiment that finds bias in images throughout their disclosure ([0082], [0093], [0104], etc.). See Tieu Figs. 8 and 9 for the way their obfuscation method integrates into the final image.
“ranking the identified protected features;” Dalli teaches “A second exemplary method makes use of local feature importance extracted from the rule-based model coefficients in order to identify any potential bias locally.” ([0011]) see [0057]-[0058] for a mapping from model coefficients to features producing a feature importance ranking. See also [0065] for feature importance being ranked. Shanbahg’s method also highlights important features, or groups of features, based on the prediction drift (Section 9.2). The Examiner contends the direct mapping that Dalli demonstrates between the model coefficients and features, in combination with the methods by which Shanbahg indicates important features, sufficiently reads on the broad limitation of “ranking…” the features.
“generating, using a generative adversarial network, adversarial artifacts to obfuscate at least a portion of the ranked protected features;” While Dalli does teach an embodiment that uses a generative model ([0183]), the combination of Dalli and Shanbahg fails to explicitly teach “generating… adversarial artifacts to obfuscate at least a portion of the ranked protected features;” However, Tieu teaches “We have developed a spatio-temporal generative adversarial network (ST-GAN) that uses random noise synthesized in the gait distribution to generate anonymized gaits that appear natural. ST-GAN consists of a generator that uses the original gait and random noise to generate an anonymized gait and two discriminators, a spatial discriminator and a temporal discriminator, to estimate the probability that a gait is the original one and not an anonymized one.” (Abstract).
“altering the background of the media by integrating the adversarial artifacts into the background of the media and by excluding the ranked protected features from the media;” Tieu teaches the addition of noise into a video of a person’s gait, with the aim of keeping the identifying information obfuscated. Tieu uses a GAN to generate noise that is integrated with the video of the person’s gait. (Page 309-310, Section 3.2). See Tieu Figs. 8 and 9 for the way their obfuscation method integrates into the final image and Section 3.5 (Pages 311-312) for how the image background is manipulated as well.
Tieu highlights that the need for the privacy protection capabilities of their system, particularly with something as personally identifiable as one’s gait (Introduction). It would have been obvious to one of ordinary skill in the art at the time of the applicant’s filing to combine the detection methods of Dalli while also obfuscating the biased or personally identifiable features similar to Tieu to improve the bias detection and mitigation of the model.
While Shanbhag makes reference to the use of Jensen Shannon divergence (Section 9.2), which uses the Kullback-Liebler divergence (Appendix 11.2), The combination of Dalli, Shanbhag, and Tieu fails to explicitly teach: “determining that a Kullback-Leibler divergence, used to evaluate detectability of the ranked protected features, is not less than a threshold; determining, based on the determining that the Kullback-Leibler divergence is not less than the threshold, that the generative adversarial network is not able to detect each of the ranked protected features in the media after integrating the adversarial artifacts into the media;” Luo, in a similar field of endeavor of data obfuscation teaches the use of a Kullback-Leibler divergence to measure the probability distributions of an original and obfuscated signals (or images). The goal of Luo’s paper is to minimize the KL measure to 0 (0 KL means the distributions are identical, and the model cannot detect the obfuscations) (Luo, Introduction, Page 1). While Luo does not teach a threshold for the KL specifically, it would have been obvious to one having ordinary skill in the art at the time the invention was made to reduce the KL measure towards 0, since it has been held that discovering an optimum value of a result effective variable involves only routine skill in the art. In re Boesch, 617 F.2d 272,265 USPQ 215 (CCPA 1980).
Luo demonstrates the Kullback-Leibler divergence would have been a well-known metric for data obfuscation to a person of ordinary skill in the art at the time of the Applicant’s filing. It would have been obvious to such a person to leverage known methods to realize known benefits in a combination of Dalli, Shanbhag, Tieu, and Luo.
“selecting, based on determining that the generative adversarial network is not able to detect each of the ranked protected features, a dynamic option or a static option for incorporating adversarial data into the media;” Based on the Applicant’s Specification [0068], the Examiner interprets that Tieu reads on this limitation by collecting/analyzing/obfuscating frames of data as well as sequences of frames of data. See Table 1 and Section 3 (Page 309).
In regards to claim 2: The present invention claims: “wherein the protected features further comprise protected attributes and protected cues.” Dalli teaches protected attributes ([0044]).
In regards to claim 3: The present invention claims: “obtaining knowledge information on the AI system and the plurality of candidate features from the media;” Dalli teaches “Methods for detecting bias, strength, and weakness of data sets and the resulting models may be described.” (Abstract) Dalli’s Abstract also outlines various ways that bias is detected either in the data or models.
“and determining a type characteristic of the model of the AI system, wherein the identifying of the protected features is based on the type characteristic of the model.” Dalli teaches “Various methods for detecting the bias and weakness (and conversely the strength) of the data sets and also the resulting models may be described. A first exemplary method presents a global bias detection which utilizes the coefficients of the XAI/XNN/XTT/XSN/XMN model to identify any potential bias. A second exemplary method makes use of local feature importance extracted from the rule-based model coefficients in order to identify any potential bias locally. A third exemplary method aggregates the feature importance over the results/explanations of multiple samples. A fourth exemplary method presents a special method for detecting bias in multi-dimensional data such as images.” ([0011], mapping the use of model coefficients or importance extracted from model rules as “type characteristic” and how those coefficients or importances can point to biased or protected features). Shanbahg’s method also highlights important features, or groups of features, based on the prediction drift (Section 9.2).
In regards to claim 7: The present invention claims: “wherein the generative adversarial network comprises a generator, a first discriminant network, and a second discriminant network, wherein the first discriminant network serves to classify the ranked protected features and constrain the media to confuse a corresponding type of protected feature of the ranked protected features, and the second discriminant network that serves to discriminate whether a given sample is real or generated,” Se Dalli Fig. 2 where data is synthesized in 202, a first network performs a prediction in 204 and 206, the partitions, transformations, and rules of 208/210/212 dictate which features gain importance and aid in the detection of bias, and the creation of a second white-box model for the final output. See [0203] where Dalli teaches an embodiment that may label or classify synthetic or sampled data. Dalli also teaches in paragraphs [0057]-[0058] a mapping between model coefficients and protected or important features. It would follow that an adjustment to model weights/coefficients (modification), and therefore a resultant change in the output of the model (prediction drift), would also result in change to the ranking of mapped features to the coefficient vector, resulting in a change to prospected features.
“a first term of a minmax function of the generative adversarial network optimizes for the integrated media to match unintegrated media, and a second term of the minmax function causes a generation of a representation of the integrated media in which the at least a portion of the ranked protected features are obfuscated.” See Tieu Section 3.4 for the method by which they maximize their noise to obfuscate the gait, while retaining as much information as possible such as viewing angle and the action being performed (Page 310-311, mapping the minimization of difference in original information to optimizing matching the media and the second term being the obfuscation of the attributes through noise).
In regards to claims 9-11 and 15: Claims 9-11 and 15 recite similar limitations to claims 1-3 and 7, with the exception of an apparatus in claim 9. Therefore, both sets of limitations are similarly rejected.
In regards to claims 17-18: Claims 17-18 recite similar limitations to claims 1 and 3, with the exception of a computer program product in claim 17. Therefore, both sets of limitations are similarly rejected.
In regards to claim 21: The present invention claims: “incorporating the adversarial data into the media in a case where the dynamic option is selected, wherein the adversarial data is different from the adversarial artifacts; and incorporating the adversarial artifacts into the media in a case where the static option is selected.” Based on the Applicant’s Specification [0068], the Examiner interprets that Tieu reads on this limitation by collecting/analyzing/obfuscating frames of data as well as sequences of frames of data. See Table 1 and Section 3 (Page 309), as Tieu modifies frames or sequences of frames as the gait proceeds through the video.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to GRIFFIN T BEAN whose telephone number is (703)756-1473. The examiner can normally be reached M - F 7:30 - 4:30.
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, Li Zhen can be reached at (571) 272-3768. 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.
/GRIFFIN TANNER BEAN/Examiner, Art Unit 2121
/Li B. Zhen/Supervisory Patent Examiner, Art Unit 2121
Prosecution Insights