Prosecution Insights
Last updated: November 12, 2026
Application No. 18/745,788

SYSTEM AND METHOD FOR DETECTING ETHICAL BIAS IN MACHINE LEARNING MODELS

Non-Final OA §101§103
Filed
Jun 17, 2024
Priority
Jun 16, 2023 — provisional 63/508,739
Examiner
BUI, BRIAN DUYQUANG
Art Unit
2127
Tech Center
2100 — Computer Architecture & Software
Assignee
Hsbc Group Management Services Limited
OA Round
1 (Non-Final)
Grant Probability
Favorable
1-2
OA Rounds

Examiner Intelligence

Grants only 0% of cases
0%
Career Allowance Rate
0 granted / 0 resolved
-55.0% vs TC avg
Minimal +0% lift
Without
With
+0.0%
Interview Lift
resolved cases with interview
Typical timeline
Avg Prosecution
9 currently pending
Career history
9
Total Applications
across all art units
This examiner has no resolved cases yet (career too new); statute-level performance unavailable. The Grant Probability card shows Tech Center averages instead.

Office Action

§101 §103
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 . Claims 1-20 are pending for examination. Claims 1, 11, 20 are independent. Priority Acknowledgement is made of applicant’s claim for domestic priority with regard to US Provisional Application No. 63/508739, entitled “System and Method for Detecting Ethical Bias in Machine Learning Models”, dated June 16, 2023. Claim Objections Claims 5-8 and 15-18 are objected to because of the following informalities: Claims 5 and 15: "upon a determination that the updated estimated bias output value and updated materiality output value . Claims 6 and 16: “re-run a bias test using bureau variables as the control variablesand generate an updated estimated bias output value” Claims 7-8 and 17-18: “generat[e/ing] one or more backup snapshot instances of the trained machine learning model.” Claim 17: “generating…” Appropriate correction is required. Claim Interpretation The following is a quotation of 35 U.S.C. 112(f): (f) Element in Claim for a Combination. – An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof. The following is a quotation of pre-AIA 35 U.S.C. 112, sixth paragraph: An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof. The claims in this application are given their broadest reasonable interpretation using the plain meaning of the claim language in light of the specification as it would be understood by one of ordinary skill in the art. The broadest reasonable interpretation of a claim element (also commonly referred to as a claim limitation) is limited by the description in the specification when 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is invoked. As explained in MPEP § 2181, subsection I, claim limitations that meet the following three-prong test will be interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph: (A) the claim limitation uses the term “means” or “step” or a term used as a substitute for “means” that is a generic placeholder (also called a nonce term or a non-structural term having no specific structural meaning) for performing the claimed function; (B) the term “means” or “step” or the generic placeholder is modified by functional language, typically, but not always linked by the transition word “for” (e.g., “means for”) or another linking word or phrase, such as “configured to” or “so that”; and (C) the term “means” or “step” or the generic placeholder is not modified by sufficient structure, material, or acts for performing the claimed function. Use of the word “means” (or “step”) in a claim with functional language creates a rebuttable presumption that the claim limitation is to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites sufficient structure, material, or acts to entirely perform the recited function. Absence of the word “means” (or “step”) in a claim creates a rebuttable presumption that the claim limitation is not to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is not interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites function without reciting sufficient structure, material or acts to entirely perform the recited function. Claim limitations in this application that use the word “means” (or “step”) are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action. Conversely, claim limitations in this application that do not use the word “means” (or “step”) are not being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action. This application includes one or more claim limitations that do not use the word “means,” but are nonetheless being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, because the claim limitation(s) uses a generic placeholder that is coupled with functional language without reciting sufficient structure to perform the recited function and the generic placeholder is not preceded by a structural modifier. Such claim limitation(s) is/are: "flag data structure" in Claims 6 and 16. [0015, 0033, 0057] of Applicant’s specification. Because this/these claim limitation(s) is/are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, it/they is/are being interpreted to cover the corresponding structure described in the specification as performing the claimed function, and equivalents thereof. If applicant does not intend to have this/these limitation(s) interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, applicant may: (1) amend the claim limitation(s) to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph (e.g., by reciting sufficient structure to perform the claimed function); or (2) present a sufficient showing that the claim limitation(s) recite(s) sufficient structure to perform the claimed function so as to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. 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. Step 1: The claim is a process, machine, manufacture, or composition of matter. In the instant application, Claims 1-10 are directed to a machine, Claims 11-19 are directed to a method, and Claim 20 is directed to a manufacture. Thus, each of the claims falls within one of the four statutory categories (i.e. process, machine, manufacture, or composition of matter). With regard to Claim 1: 2A Prong 1: The claim recites an abstract idea, law of nature, or natural phenomenon. identify a bias variable for testing the target trained machine learning model; (This step of identifying a variable to test a machine learning model is practically performable in the human mind and is understood to be a recitation of a mental process with the aid of pen and paper (i.e. observation).) generate a first data set, the first data set comprising a set of control variables and bias variables for input to the target trained machine learning model; (This step of generating a dataset is practically performable in the human mind and is understood to be a recitation of a mental process with the aid of pen and paper (i.e. observation).) generate a second data set, the second data set comprising the set of control variables, and not the bias variables, for input to the trained target machine learning model; (This step of generating a dataset is practically performable in the human mind and is understood to be a recitation of a mental process with the aid of pen and paper (i.e. observation).) process, the first data set and the second data set (This step of processing information to generate an output is practically performable in the human mind and is understood to be a recitation of a mental process with the aid of pen and paper (i.e. judgement).) determine an estimated bias output value based at least on a fit of a linear model between the first output data and the second output data; (This step of identifying a variable to test a machine learning model is practically performable in the human mind and is understood to be a recitation of a mental process with the aid of pen and paper (i.e. judgement).) determine a materiality output value based at least on a likelihood ratio test conducted against the estimated bias output value; and (This step of identifying a variable to test a machine learning model is practically performable in the human mind and is understood to be a recitation of a mental process with the aid of pen and paper (i.e. judgement).) 2A Prong 2: The judicial exception is not integrated into a practical application. A computer implemented system for computationally analyzing a target trained machine learning model to generate at least one estimation value representative of a potential ethical bias present in the target trained machine learning model, the system comprising: a computer processor operating in conjunction with non-transitory computer readable memory, the computer processor configured to: (The processor and non-transitory computer readable memory are understood as mere instructions to apply the exception using a generic computer component — see MPEP 2106.05(f).) process, the first data set and the second data set using the trained target machine learning model to generate a corresponding first set of output data and a second set of output data; (This step is reciting a judicial exception with the words “apply it” (or an equivalent), or merely invoking computers or machinery as a tool to perform the abstract idea (i.e. evaluation) — see MPEP 2106.05(f).) record at least one of the estimated bias output value and the materiality output value in a data storage as metadata associated with the target trained machine learning model. (This step is reciting a judicial exception with the words “apply it” (or an equivalent), or merely invoking computers or machinery as a tool to perform the abstract idea (i.e. observation) — see MPEP 2106.05(f).) The additional elements as disclosed above alone or in combination do not integrate the judicial exception into practical application as they are merely invoking computers as a tool to perform the abstract idea. 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. A computer implemented system for computationally analyzing a target trained machine learning model to generate at least one estimation value representative of a potential ethical bias present in the target trained machine learning model, the system comprising: a computer processor operating in conjunction with non-transitory computer readable memory, the computer processor configured to: (The processor and non-transitory computer readable memory are understood as mere instructions to apply the exception using a generic computer component — see MPEP 2106.05(f).) process, the first data set and the second data set using the trained target machine learning model to (This step is reciting a judicial exception with the words “apply it” (or an equivalent), or merely invoking computers or machinery as a tool to perform the abstract idea (i.e. evaluation) — see MPEP 2106.05(f).) record at least one of the estimated bias output value and the materiality output value in a data storage as metadata associated with the target trained machine learning model. (Recording information is understood as an insignificant extra-solution activity, being exemplary of selecting information for collection, analysis and display — see MPEP 2106.05(g).) The additional elements as disclosed above alone or in combination do not recite significantly more than a judicial exception as they are merely invoking computers as a tool to perform the abstract idea. With regard to Claim 11: Claim 11 is a method claim that has similar limitations to Claim 1; therefore, it is rejected under the same rationale. With regard to Claim 20: Claim 20 is a non-transitory computer readable medium claim that has similar limitations to Claim 1; therefore, it is rejected under the same rationale. Additional limitations of Claim 20 are addressed below. 2A Prong 2 & 2B: A non-transitory computer readable medium, storing machine interpretable instructions, which when executed by a processor, cause the processor to perform a computer implemented method for computationally analyzing a target trained machine learning model to generate at least one estimation value representative of a potential ethical bias present in the target trained machine learning model, the method comprising: (The non-transitory computer readable medium is understood as mere instructions to apply the exception using a generic computer component — see MPEP 2106.05(f).) With regard to Claims 2 and 12: 2A Prong 1: The claim recites an abstract idea, law of nature, or natural phenomenon. The computer implemented system of claim 1 and the computer implemented method of claim 11 respectively, wherein the computer processor is further configured to: upon a determination that the estimated bias output value and the materiality output value associated with the target trained machine learning model are both greater than a threshold data value, (This step of determining whether a value is greater than a threshold is practically performable in the human mind and is understood to be a recitation of a mental process with the aid of pen and paper (i.e. evaluation).) 2A Prong 2: The judicial exception is not integrated into a practical application. The computer implemented system of claim 1 and the computer implemented method of claim 11 respectively, wherein the computer processor is further configured to: (This step of generating is reciting a judicial exception with the words “apply it” (or an equivalent), or merely invoking computers or machinery as a tool to perform the abstract idea (i.e. evaluation) — see MPEP 2106.05(f).) and transmit an instruction signal to a model pipeline control data process to cease usage of the target trained machine learning model. (Transmitting information is understood as an insignificant extra-solution activity — see MPEP 2106.05(g).) 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. The computer implemented system of claim 1 and the computer implemented method of claim 11 respectively, wherein the computer processor is further configured to: (This step of generating is reciting a judicial exception with the words “apply it” (or an equivalent), or merely invoking computers or machinery as a tool to perform the abstract idea (i.e. evaluation) — see MPEP 2106.05(f).) and transmit an instruction signal to a model pipeline control data process to cease usage of the target trained machine learning model. (Transmitting information is understood as a well-understood, routine, and conventional function, being exemplary of receiving or transmitting data over a network — see MPEP 2106.05(d)(II)(i).) With regard to Claims 3 and 13: 2A Prong 1: The claim recites an abstract idea, law of nature, or natural phenomenon. The computer implemented system of claim 2 and the computer implemented method of claim 12 respectively, wherein the computer processor is further configured to: identify a recent past instance of the target trained machine learning model; and (This step of identifying a recent past instance of the machine learning model is practically performable in the human mind and is understood to be a recitation of a mental process with the aid of pen and paper (i.e. observation).) 2A Prong 2: The judicial exception is not integrated into a practical application. generate (This step of generating is reciting a judicial exception with the words “apply it” (or an equivalent), or merely invoking computers or machinery as a tool to perform the abstract idea (i.e. evaluation) — see MPEP 2106.05(f).) and transmit an instruction signal to the model pipeline control data process to begin usage of the recent past instance of the target trained machine learning model. (Transmitting information is understood as an insignificant extra-solution activity — see MPEP 2106.05(g).) 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. generate (This step of generating is reciting a judicial exception with the words “apply it” (or an equivalent), or merely invoking computers or machinery as a tool to perform the abstract idea (i.e. evaluation) — see MPEP 2106.05(f).) and transmit an instruction signal to the model pipeline control data process to begin usage of the recent past instance of the target trained machine learning model. (Transmitting information is understood as a well-understood, routine, and conventional function, being exemplary of receiving or transmitting data over a network — see MPEP 2106.05(d)(II)(i).) With regard to Claims 4 and 14: 2A Prong 1: The claim recites an abstract idea, law of nature, or natural phenomenon. The computer implemented system of claim 2 and the computer implemented method of claim 12 respectively, wherein the computer processor is further configured to: identify an alternate trained machine learning model; and (This step of identifying an alternate machine learning model is practically performable in the human mind and is understood to be a recitation of a mental process with the aid of pen and paper (i.e. observation).) 2A Prong 2: The judicial exception is not integrated into a practical application. generate and transmit an instruction signal to the model pipeline control data process to begin usage of the alternate trained machine learning model. (This step of generating is reciting a judicial exception with the words “apply it” (or an equivalent), or merely invoking computers or machinery as a tool to perform the abstract idea (i.e. evaluation) — see MPEP 2106.05(f).) (Transmitting information is understood as an insignificant extra-solution activity — see MPEP 2106.05(g).) 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. generate and transmit an instruction signal to the model pipeline control data process to begin usage of the alternate trained machine learning model. (This step of generating is reciting a judicial exception with the words “apply it” (or an equivalent), or merely invoking computers or machinery as a tool to perform the abstract idea (i.e. evaluation) — see MPEP 2106.05(f).) (Transmitting information is understood as a well-understood, routine, and conventional function, being exemplary of receiving or transmitting data over a network — see MPEP 2106.05(d)(II)(i).) With regard to Claims 5 and 15: 2A Prong 1: The claim recites an abstract idea, law of nature, or natural phenomenon. determine an updated estimated bias output value based at least on the fit of the linear model between outputs of the first data set and the second data set run on a re-trained target trained machine learning model; (This step of determining an output value is practically performable in the human mind and is understood to be a recitation of a mental process with the aid of pen and paper (i.e. judgement).) determine an updated materiality output value based at least on a likelihood ratio test conducted against the estimated bias output value; and (This step of determining an output value is practically performable in the human mind and is understood to be a recitation of a mental process with the aid of pen and paper (i.e. judgement).) upon a determination that the updated estimated bias output value and updated materiality output value being below the threshold data value; (This step of determining a whether a value is below a threshold value is practically performable in the human mind and is understood to be a recitation of a mental process with the aid of pen and paper (i.e. evaluation).) 2A Prong 2: The judicial exception is not integrated into a practical application. The computer implemented system of claim 2 and the computer implemented method of claim 12 respectively, wherein the computer processor is further configured to: generate (This step of generating is reciting a judicial exception with the words “apply it” (or an equivalent), or merely invoking computers or machinery as a tool to perform the abstract idea (i.e. evaluation) — see MPEP 2106.05(f).) and transmit an instruction signal to the model pipeline control data process to initiate retraining of the target trained machine learning model; (Transmitting information is understood as an insignificant extra-solution activity — see MPEP 2106.05(g).) (This step of generating is reciting a judicial exception with the words “apply it” (or an equivalent), or merely invoking computers or machinery as a tool to perform the abstract idea (i.e. evaluation) — see MPEP 2106.05(f).) and transmit an instruction signal to the model pipeline control data process to reinstate the re-trained target trained machine learning model. (Transmitting information is understood as an insignificant extra-solution activity — see MPEP 2106.05(g).) 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. The computer implemented system of claim 2, wherein the computer processor is further configured to: generate (This step of generating is reciting a judicial exception with the words “apply it” (or an equivalent), or merely invoking computers or machinery as a tool to perform the abstract idea (i.e. evaluation) — see MPEP 2106.05(f).) and transmit an instruction signal to the model pipeline control data process to initiate retraining of the target trained machine learning model; (Transmitting information is understood as a well-understood, routine, and conventional function, being exemplary of receiving or transmitting data over a network — see MPEP 2106.05(d)(II)(i).) (This step of generating is reciting a judicial exception with the words “apply it” (or an equivalent), or merely invoking computers or machinery as a tool to perform the abstract idea (i.e. evaluation) — see MPEP 2106.05(f).) and transmit an instruction signal to the model pipeline control data process to reinstate the re-trained target trained machine learning model. (Transmitting information is understood as a well-understood, routine, and conventional function, being exemplary of receiving or transmitting data over a network — see MPEP 2106.05(d)(II)(i).) With regard to Claims 6 and 16: 2A Prong 1: The claim recites an abstract idea, law of nature, or natural phenomenon. The computer implemented system of claim 2 and the computer implemented method of claim 12 respectively, wherein the computer processor is further configured to: upon a determination that the estimated bias output value and the materiality output value associated with the target trained machine learning model are both greater than a threshold data value, (This step of determining whether a value is greater than a threshold value is practically performable in the human mind and is understood to be a recitation of a mental process with the aid of pen and paper (i.e. evaluation).) upon a second determination that the updated estimated bias output value and the updated materiality output value associated with the target trained machine learning model are below the threshold data value (This step of determining whether a value is below a threshold value is practically performable in the human mind and is understood to be a recitation of a mental process with the aid of pen and paper (i.e. evaluation).) 2A Prong 2 & 2B: The computer implemented system of claim 2 and the computer implemented method of claim 12 respectively, wherein the computer processor is further configured to: (This step of re-running a test is reciting a judicial exception with the words “apply it” (or an equivalent), or merely invoking computers or machinery as a tool to perform the abstract idea (i.e. evaluation) — see MPEP 2106.05(f).) generating an updated estimated bias output value and updated materiality output value; and (This step of generating is reciting a judicial exception with the words “apply it” (or an equivalent), or merely invoking computers or machinery as a tool to perform the abstract idea (i.e. evaluation) — see MPEP 2106.05(f).) associate a flag data structure with the target trained machine learning model during continued operation of the target trained machine learning model. (This step of associating a data structure with the machine learning model is reciting a judicial exception with the words “apply it” (or an equivalent), or merely invoking computers or machinery as a tool to perform the abstract idea (i.e. evaluation) — see MPEP 2106.05(f).) With regard to Claims 7 and 17: 2A Prong 1: The claim recites an abstract idea, law of nature, or natural phenomenon. The computer implemented system of claim 1 and the computer implemented method of claim 11 respectively, wherein the computer processor is further configured to: periodically determine and record time-stamped versions of the estimated bias output value and the materiality output value for the trained machine learning model; and (This step of determining and recording time-stamped versions of the values is practically performable in the human mind and is understood to be a recitation of a mental process with the aid of pen and paper (i.e. determining is judgement, recording is observation).) 2A Prong 2 & 2B: generate one or more backup snapshot instances of the trained machine learning, triggered by the estimated bias output value and the materiality output value being below a threshold data value; wherein the trained machine learning model is a reinforcement learning model. (This step is reciting a judicial exception with the words “apply it” (or an equivalent), or merely invoking computers or machinery as a tool to perform the abstract idea (i.e. evaluation) — see MPEP 2106.05(f).) With regard to Claims 8 and 18: 2A Prong 1: The claim recites an abstract idea, law of nature, or natural phenomenon. The computer implemented system of claim 1 and the computer implemented method of claim 11 respectively, wherein the computer processor is further configured to: periodically determine and record time-stamped versions of the estimated bias output value and the materiality output value for the trained machine learning model; and (This step of determining and recording time-stamped versions of the values is practically performable in the human mind and is understood to be a recitation of a mental process with the aid of pen and paper (i.e. determining is judgement, recording is observation).) 2A Prong 2 & 2B: generate one or more backup snapshot instances of the trained machine learning, triggered by the estimated bias output value and the materiality output value being below a threshold data value; wherein the trained machine learning model is a reinforcement learning model. (This step of generating is reciting a judicial exception with the words “apply it” (or an equivalent), or merely invoking computers or machinery as a tool to perform the abstract idea (i.e. evaluation) — see MPEP 2106.05(f).) 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. Claims 1, 11, and 20 are rejected under 35 U.S.C. §103 as being unpatentable over Hacmon et al. (US 2022/0076080 A1), hereinafter “Hacmon”, in view of Elovici et al. (US 2021/0357508 A1), hereinafter “Elovici”. With regard to Claim 1: Hacmon teaches: A computer implemented system for computationally analyzing a target trained machine learning model to generate at least one estimation value representative of a potential ethical bias present in the target trained machine learning model, the system comprising: ([0033-0035] Hacmon describes the fairness score determined from the tests performed to determine a level of bias “in the examined ML model.” [0039] “The system performs different tests to examine the ML model’s fairness levels.”) a computer processor operating in conjunction with non-transitory computer readable memory, the computer processor configured to: ([0038] Hacmon describes the invention as a “system (plugin)” for assessment of bias in Artificial Intelligence based machine learning models , which implies the use of processor and memory to obtain results from the model(s).) identify a bias variable for testing the target trained machine learning model; ([0047, 0050] Hacmon states that the data/model profiler of the system gathers data properties such as protected features to guide the test execution environment (i.e., testing the target trained machine learning model). [0050] Hacmon states “in order to perform the bias estimation [i.e., testing the target trained machine learning model], the protected feature should be defined [i.e., identify a bias variable]”.) generate a first data set, the first data set comprising a set of control variables and bias variables for input to the target trained machine learning model; ([0086] Hacmon provides an example of datasets to be tested “”no disability” vs. not “no disability””, wherein “not “no disability”” is understood as a dataset comprised of the control variables and the bias variables (e.g. disability). [0106-0109] Hacmon describes equations that use datasets that contain both protected class (bias) labels fp and true (control) labels yt.) generate a second data set, the second data set comprising the set of control variables, and not the bias variables, for input to the trained target machine learning model; ([[0086] Hacmon provides an example of datasets to be tested “”no disability” vs. not “no disability””, wherein “no disability” is understood as a dataset comprised of just the control variables and not the bias variables (e.g. disability). [0106-0109] Hacmon describes equations that use datasets that contain just the true (control) labels yt without the protected (bias) labels fp.) process, the first data set and the second data set using the trained target machine learning model to generate a corresponding first set of output data and a second set of output data; ([0106-0109] Hacmon describes the “equalized odds” and “equal opportunity” tests that examine the dataset without the protected (bias) label fp against the dataset that contains both the true (control) label yt and the protected label, wherein the model prediction y for each dataset is understood as the corresponding first and second output data.) determine an estimated bias output value based at least on a fit of a linear model between the first output data and the second output data; ([0054] Hacmon recites “Each test determines whether underlying bias exists in the ML model. [0106-0109] Hacmon describes the “equalized odds” and “equal opportunity” tests that examine the two probabilities (i.e. first and second output data), and “produce a single scaled fairness score result” labeled “result” which is understood as an estimated bias output value. [0107, 0109] The aforementioned tests examine the variance in predictions y between each dataset, in which the analysis of variance is understood as a general linear model.) determine a materiality output value based at least on a likelihood ratio test conducted against the estimated bias output value; ([0054-0060, 0096-0097] Hacmon discusses the use of the likelihood ratio positive test for the protected (i.e. bias) feature group that examines the ratio between sensitivity and its complement value, wherein the sensitivity result is understood as a bias output value and the output of the likelihood ratio test is understood as a materiality output.) Hacmon does not explicitly teach: record at least one of the estimated bias output value and the materiality output value in a data storage as metadata associated with the target trained machine learning model. However, Elovici teaches in the same field of endeavor: record at least one of the estimated bias output value and the materiality output value in a data storage as metadata associated with the target trained machine learning model. ([0090-0091] Elovici describes storage of the metadata of tests that test machine learning models for bias. [0101] Elovici clarifies that test scores and results are attached to each test. [Fig. 4] Elovici provides an example interface of tests with final scores associated with a target trained machine learning model.) It would have been obvious for one of ordinary skill in the art prior to the effective filing date of the claimed invention to have modified Hacmon’s teachings by storing the test results indicative of bias as metadata as taught by Elovici. One would have been motivated to make this modification in order to allow users to more easily access bias indication data for models to be used in adjusting or selecting machine learning models for implementation. With respect to Claim 11: Hacmon in view of Elovici teaches: A computer implemented method for (Claim 11 is a method claim that has similar limitations to Claim 1; therefore, it is rejected under the same rationale.) With respect to Claim 20: Hacmon in view of Elovici teaches: A non-transitory computer readable medium, storing machine interpretable instructions, which when executed by a processor, cause the processor to perform a computer implemented method for (Claim 20 is a non-transitory computer readable medium claim that has similar limitations to Claim 1; therefore, it is rejected under the same rationale.) Claims 2-5 and 12-15 are rejected under 35 U.S.C. §103 as being unpatentable over Hacmon in view of Elovici, and further in view of Chalamalasetti et al. (US 2022/0121885 A1), hereinafter “Chalamalasetti”, and Wang et al. (US 2025/0048140 A1), hereinafter “Wang”. With respect to Claims 2 and 12: Hacmon in view of Elovici teaches: The computer implemented system of claim 1 and the computer implemented method of claim 11 respectively, wherein the computer processor is further configured to: upon a determination that the estimated bias output value and the materiality output value associated with the target trained machine learning model ([0054] Hacmon explains that “each test determines whether underlying bias exists in the ML model [i.e., the target trained machine learning model].” [0058] Hacmon teaches the sensitivity test that determines a bias output value. [0060] Hacmon teaches the likelihood ratio positive test that determines an output (i.e., materiality output value) based on the bias output value from the sensitivity test.) Hacmon in view of Elovici does not explicitly teach: upon a determination that the estimated bias output value and the materiality output value associated with the target trained machine learning model are both greater than a threshold data value, generate and transmit an instruction signal to a model pipeline control data process to cease usage of the target trained machine learning model. However, Chalamalasetti teaches in the same field of endeavor: upon a determination that the estimated bias output value and the ([0043] Chalamalasetti states “a negative determination (i.e., presence of bias) at block 414 may require at least one accumulative metric (i.e., bias output value) exceeding a corresponding threshold value and at least one standard deviation metric exceeding a corresponding threshold value.”) It would have been obvious for one of ordinary skill in the art prior to the effective filing date of the claimed invention to have modified Hacmon in view of Elovici’s teachings by comparing both the bias output value and a statistical value related to the materiality or confidence of the bias value to a threshold value as taught by Chalamalasetti. One would have been motivated to make this modification in order to further ensure that the bias detection is statistically significant or material enough to give further attention. This allows for increased accuracy of predictions in allowing for non-significant bias variables to remain in the data and provide a wider range of data for model training and testing. Hacmon in view of Elovici and Chalamalasetti does not explicitly teach: upon a determination that the estimated bias output value and the materiality output value associated with the target trained machine learning model are both greater than a threshold data value, generate and transmit an instruction signal to a model pipeline control data process to cease usage of the target trained machine learning model. However, Wang teaches in the same field of endeavor: upon a determination that ([0070] Wang teaches making a determination to “stop the AI/ML model… in case of poor performance (a performance result is lower than a performance threshold).” [0072] Wang teaches the first network device transmitting indication information (i.e., instruction signal) to the terminal equipment (i.e., model pipeline control data process) for stopping (i.e., cease usage) an AI/ML model. [0079] Wang further clarifies that the indication information is carried by an XnAP message which may be a new (i.e., generated) XnAP message.) It would have been obvious for one of ordinary skill in the art prior to the effective filing date of the claimed invention to have modified Hacmon in view of Elovici and Chalamalasetti’s teachings by generating and transmitting a signal to stop the ML model upon threshold detection as taught by Wang. One would have been motivated to make this modification in order to prevent further use of a biased and unfair model to process information in real-world situations such as credit lending from financial institutions. With respect to Claims 3 and 13: Hacmon in view of Elovici and Chalamalasetti teaches: The computer implemented system of claim 2 and the computer implemented method of claim 12 respectively, wherein the computer processor is further configured to: identify a recent past instance of the target trained machine learning model; and ([Fig. 3, 0079] Elovici describes models stored in a Secure Data Repository for further reuse (i.e., begin usage of the past instance of the ML model). Elovici further recites “each change to the ML model creates a new (i.e., recent) model version… Each model version is associated with its parent model [i.e., past instances of the target trained machine learning model], for quick retrieval (i.e., the instance is identified), in case of required revisions.”) Hacmon in view of Elovici and Chalamalasetti does not explicitly teach: identify a recent past instance of the target trained machine learning model; and generate and transmit an instruction signal to the model pipeline control data process to begin usage of the recent past instance of the target trained machine learning model. However, Wang teaches in the same field of endeavor: identify a ([0080] Wang teaches the indication information (i.e., identification) including version (i.e., instance) information of the AI/ML model.) and generate and transmit an instruction signal to the model pipeline control data process to begin usage of the ([0072] Wang teaches the first network device transmitting indication information (i.e., instruction signal) to the terminal equipment (i.e., model pipeline control data process) for updating an AI/ML model. [0103] Wang teaches a lower version model being updated to a higher version model (i.e., begin usage of an instance of the target ML model). [0079] Wang further clarifies that the indication information is carried by an XnAP message which may be a new (i.e., generated) XnAP message.) It would have been obvious for one of ordinary skill in the art prior to the effective filing date of the claimed invention to have modified Hacmon in view of Elovici and Chalamalasetti’s teachings by generating and transmitting a signal to switch the ML model to a previous instance as taught by Wang. One would have been motivated to make this modification in order to ensure that the model in use has been previously verified to be within acceptable limits of bias and allows the data processing to continue in real time when the target ML model is deemed to be too unfair to use. With respect to Claims 4 and 14: Hacmon in view of Elovici and Chalamalasetti teaches: The computer implemented system of claim 2 and the computer implemented method of claim 12 respectively, wherein the computer processor is further configured to: identify an alternate trained machine learning model; ([Fig. 3, 0076] Elovici explains that multiple models (“Machine Learning Model n”) may be continuously tested for robustness in the project, indicating that the multiple (i.e., alternate) trained machine learning models are identified in order to be tested.) Hacmon in view of Elovici and Chalamalasetti does not explicitly teach: identify an alternate trained machine learning model; and generate and transmit an instruction signal to the model pipeline control data process to begin usage of the alternate trained machine learning model. However, Wang teaches in the same field of endeavor: identify an alternate trained machine learning model; and generate and transmit an instruction signal to the model pipeline control data process to begin usage of the alternate trained machine learning model. ([[0072] Wang teaches the first network device transmitting indication information (i.e., instruction signal) to the terminal equipment (i.e., model pipeline control data process) for switching of an AI/ML model. [0198] Wang clarifies that the terminal equipment may make a decision to switch the model to another model (i.e., an alternate trained machine learning model). [0079] Wang further clarifies that the indication information is carried by an XnAP message which may be a new (i.e., generated) XnAP message.) It would have been obvious for one of ordinary skill in the art prior to the effective filing date of the claimed invention to have modified Hacmon in view of Elovici and Chalamalasetti’s teachings by generating and transmitting a signal to switch the ML model to another alternative model as taught by Wang. One would have been motivated to make this modification in order to allow the data processing to continue in real time when the target ML model is deemed to be too unfair to use. With respect to Claims 5 and 15: Hacmon teaches: The computer implemented system of claim 2 and the computer implemented method of claim 12 respectively, wherein the computer processor is further configured to: determine an ([0054] Hacmon recites “Each test determines whether underlying bias exists in the ML model. [0106-0109] Hacmon describes the “equalized odds” and “equal opportunity” tests that examine the two probabilities (i.e. outputs of the first and second data sets), and produce a fairness score result (i.e., an estimated bias output value). [0107, 0109] The aforementioned tests examine the variance in predictions y between each dataset, in which the analysis of variance is understood as a general linear model.) determine an ([0054] Hacmon recites “Each test determines whether underlying bias exists in the ML model. [0060] Hacmon teaches the likelihood ratio positive test that determines an output (i.e., materiality output value) based on the estimated bias output value from the sensitivity test.) ([0058] Hacmon teaches the sensitivity test that determines a bias output value. [0060] Hacmon teaches the likelihood ratio positive test that determines an output (i.e., materiality output value) based on the bias output value from the sensitivity test.) Hacmon does not explicitly teach: generate and transmit an instruction signal to the model pipeline control data process to initiate retraining of the target trained machine learning model; determine an updated estimated bias output value based at least on the fit of the linear model between outputs of the first data set and the second data set run on a determine an updated materiality output value based at least on a likelihood ratio test conducted against the estimated bias output value; and upon a determination that the updated estimated bias output value and updated materiality output value being below the threshold data value; and generate and transmit an instruction signal to the model pipeline control data process to reinstate the re-trained target trained machine learning model. However, Elovici teaches in the same field of endeavor: upon a determination ([0071] Elovici teaches revision (re-training) of the models and re-certification until the models pass all robustness tests. [0077] Elovici teaches a production state in which the re-trained models are determined to be accepted and certified, and deployed to service in a production environment (i.e., reinstate the re-trained model).) It would have been obvious for one of ordinary skill in the art prior to the effective filing date of the claimed invention to have modified Hacmon’s teachings by redeploying a re-trained ML model after testing it for robustness (bias) as taught by Elovici. One would have been motivated to make this modification in order to ensure that the re-trained model falls within an acceptable range of bias prior to deployment, else the redundancy of deploying an unfair ML model occurs. Hacmon in view of Elovici does not explicitly teach: generate and transmit an instruction signal to the model pipeline control data process to initiate retraining of the target trained machine learning model; determine an updated estimated bias output value based at least on the fit of the linear model between outputs of the first data set and the second data set run on a determine an updated materiality output value based at least on a likelihood ratio test conducted against the estimated bias output value; and upon a determination that the updated estimated bias output value and updated materiality output value being below the threshold data value; and generate and transmit an instruction signal to the model pipeline control data process to reinstate the re-trained target trained machine learning model. However, Chalamalasetti teaches in the same field of endeavor: generate and transmit an instruction signal to the model pipeline control data process to initiate retraining of the target trained machine learning model; ([0025] Chalamalasetti teaches the bias cost function enforcement engine requesting (i.e., generating and transmitting an instruction signal) the bias mitigation engine (i.e., model pipeline control data process) to initiate re-training of the target trained machine learning model.) determine an updated estimated bias output value based at least on([Fig. 5: 502-506; 0050] Chalamalasetti teaches re-testing the re-trained target machine learning model (Fig. 5: 510) using updated bias test data, resulting in an updated bias output value such as [0039] average bias in the ML model.) determine an updated ([Fig: 5: 504-506] Chalamalasetti teaches that the detected biases may include statistical metrics (i.e., output value) representative of the detected biases. [Eq. 7; 0043] Chalamalasetti shows that the statistical metric of standard deviation is conducted against the average bias result (i.e., estimated bias output value). When calculated using the updated bias output value, the statistical metric is considered an updated output value.) ([0042] Chalamalasetti teaches “determining whether the level of bias detected in the ML model 104 – as represented by one or more of the aforementioned statistical metrics -- is below a threshold level of bias.” Chalamalasetti clarifies that the statistical metrics may include [0039] average bias in the ML model (i.e., estimated bias output value) and [0041] standard deviation of bias.) ([0031] Chalamalasetti teaches redeploying (i.e., reinstate) the re-trained target trained machine learning model.) It would have been obvious for one of ordinary skill in the art prior to the effective filing date of the claimed invention to have modified Hacmon in view of Elovici’s teachings by re-training and re-deploying the unsatisfactory ML model, and retesting the model to produce updated bias and statistical values to be compared to a threshold as taught by Chalamalasetti. One would have been motivated to make this modification in order to doubly ensure that not only is bias detected by way of exceeding an acceptable threshold, but that the detected bias also exceeds a threshold level of deviation from an expected or acceptable bias, resulting in a model that may retain an acceptable level of bias to support prediction accuracy. Hacmon in view of Elovici and Chalamalasetti does not explicitly teach: generate and transmit an instruction signal to the model pipeline control data process to initiate retraining of the target trained machine learning model; upon a determination that the updated estimated bias output value and updated materiality output value being below the threshold data value; and generate and transmit an instruction signal to the model pipeline control data process to reinstate the re-trained target trained machine learning model. However, Wang teaches in the same field of endeavor: generate and transmit an instruction signal to the model pipeline control data process ([0072] Wang teaches the first network device transmitting indication information (i.e., instruction signal) to the terminal equipment (i.e., model pipeline control data process) to update the AI/ML model. [0079] Wang further clarifies that the indication information is carried by an XnAP message which may be a new (i.e., generated) XnAP message.) upon a determination ([0072] Wang teaches the first network device transmitting indication information (i.e., instruction signal) to the terminal equipment (i.e., model pipeline control data process) for updating an AI/ML model. [0079] Wang further clarifies that the indication information is carried by an XnAP message which may be a new (i.e., generated) XnAP message.) It would have been obvious for one of ordinary skill in the art prior to the effective filing date of the claimed invention to have modified Hacmon in view of Elovici and Chalamalasetti’s teachings by generating and transmitting a signal to switch the ML model to another alternative model as taught by Wang. One would have been motivated to make this modification in order to allow the data processing to continue in real time when the target ML model is deemed to be too unfair to use. Claims 6 and 16 are rejected under 35 U.S.C. §103 as being unpatentable over Hacmon in view of Elovici, Chalamalasetti, Wang, and further in view of Abrahams et al. (US 2007/0055619 A1), hereinafter “Abrahams”. With respect to Claims 6 and 16: Hacmon in view of Wang teaches: The computer implemented system of claim 2 and the computer implemented method of claim 12 respectively, wherein the computer processor is further configured to: ([0054] Hacmon explains that “each test determines whether underlying bias exists in the ML model [i.e., bias test].” [0106-0107] Hacmon describes the “equalized odds” test (i.e., bias test) that uses equations that use true labels (i.e., control variables) yt.) and generating an ([0058] Hacmon teaches the sensitivity test that determines a bias output value. [0060] Hacmon teaches the likelihood ratio positive test that determines an output (i.e., materiality output value) based on the bias output value from the sensitivity test.) Hacmon in view of Wang does not explicitly teach: upon a determination that the estimated bias output value and the materiality output value associated with the target trained machine learning model are both greater than a threshold data value, re-run a bias test using bureau variables as the control variables, and generating an updated estimated bias output value and updated materiality output value; and upon a second determination that the updated estimated bias output value and the updated materiality output value associated with the target trained machine learning model are below the threshold data value; associate a flag data structure with the target trained machine learning model during continued operation of the target trained machine learning model. However, Elovici teaches in the same field of endeavor: upon a determination ([0104] Elovici states that in case of failed tests (i.e. bias is detected), the system may enable “re-running the last failed test suite for conforming the increase in the model resilience with respect to the tested robustness category (i.e., bias test).” [Fig. 2: 208; 0093] Elovici lists the step “manage previous similar tests and monitor changes in performance over time,” wherein monitoring changes in performance indicates an updated model performance score (i.e., updated output value) generated by the test for comparison to previous test scores. [Fig. 5] Elovici displays more than one model performance score (i.e., more than one generated output value) for the test suite.) upon a second determination that the updated estimated bias output valu([Fig. 2: 208; 0093] Elovici lists the step “manage previous similar tests and monitor changes in performance over time,” wherein monitoring changes in performance indicates an updated performance score generated by the test for comparison to previous test scores. [Fig. 5] Elovici displays more than one model score (generated output value) for the test suite.) These updated scores may be obtained after [0104] re-running the last failed test suite (i.e., a second determination).) associate a flag data structure with the target trained machine learning model during continued operation of the target trained machine learning model. (Under U.S.C. 112(f) as discussed above, “flag data structure” is given Broadest Reasonable Interpretation in light of Applicant specification [0015] “the flag can be represented in a data storage as a metadata object, value, or set of values corresponding to the model or model version.” [Fig. 5] Elovici illustrates “Status: Your model is ROBUST to abuse!” alongside “Total Score (%)” which is understood as a value associated with the target model (i.e., flag data structure) indicating a determination that the model has an acceptable level of bias. [0075] Elovici explains the Production state of a project model wherein “the models are deployed to service in a production environment (i.e., continued operation of the target trained machine learning model)”. [0094-0099] Elovici describes a generated report wherein the contents include the project state (i.e., Production state) and Project Total Score (i.e., flag).) It would have been obvious for one of ordinary skill in the art prior to the effective filing date of the claimed invention to have modified Hacmon in view of Wang’s teachings by re-running the bias test with Hacmon’s bureau control variables to obtain updated bias and materiality output values and to flag a model determined to have an acceptable level of bias as taught by Elovici. One would have been motivated to make this modification in order to determine whether the model may at least fall within an acceptable range of bias detection as predetermined by the bureau or financial institute, even when there is some level of bias detected, and to make note of this acceptable model using a flag metadata/value in order to track changes between this model and future adjusted models. Hacmon in view of Elovici and Wang does not explicitly teach: upon a determination that the estimated bias output value and the materiality output value associated with the target trained machine learning model are both greater than a threshold data value, re-run a bias test using bureau variables as the control variables, and generating an updated estimated bias output value and updated materiality output value; and upon a second determination that the updated estimated bias output value and the updated materiality output value associated with the target trained machine learning model are below the threshold data value; associate a flag data structure with the target trained machine learning model during continued operation of the target trained machine learning model. However, Chalamalasetti teaches in the same field of endeavor: upon a determination that the estimated bias output value and the ([0043] Chalamalasetti states “[i]n some example embodiments, a negative determination (i.e., presence of bias) at block 414 may require at least one accumulative metric (i.e., bias output value) exceeding a corresponding threshold value and at least one standard deviation metric exceeding a corresponding threshold value.”) upon a second determination that the updated estimated bias output value and the updated([0050] Chalamalasetti states that the ML models may be re-tested for bias (i.e., a second determination of bias), resulting in updated bias metrics.) are below the threshold data value; ([0042] Chalamalasetti teaches “determining whether the level of bias detected in the ML model 104 – as represented by one or more of the aforementioned statistical metrics -- is below a threshold level of bias.” Chalamalasetti clarifies that the statistical metrics may include [0039] average bias in the ML model (i.e., estimated bias output value) and [0041] standard deviation of bias.) It would have been obvious for one of ordinary skill in the art prior to the effective filing date of the claimed invention to have modified Hacmon in view of Elovici and Wang’s teachings by comparing both the bias output value and a statistical value related to the materiality or confidence of the bias value to a threshold value as taught by Chalamalasetti. One would have been motivated to make this modification in order to further ensure that the bias detection is statistically significant or material enough to give further attention. This allows for increased accuracy of predictions in allowing for non-significant bias variables to remain in the data and provide a wider range of data for model training and testing. Hacmon in view of Elovici, Chalamalasetti, and Wang does not explicitly teach: upon a determination that the estimated bias output value and the materiality output value associated with the target trained machine learning model are both greater than a threshold data value, re-run a bias test using bureau variables as the control variables, and generating an updated estimated bias output value and updated materiality output value; and However, Abrahams teaches in the same field of endeavor: ([0048] Abrahams teaches using credit bureau data as input data for a model used to analyze disparate treatment (i.e., bias test) in financial transactions.) It would have been obvious for one of ordinary skill in the art prior to the effective filing date of the claimed invention to have modified Hacmon in view of Elovici, Chalamalasetti, and Wang’s teachings by using bureau variables as control variables for testing the ML model as taught by Abrahams. One would have been motivated to make this modification in order to test for sensitive attributes, or variables that are not explicitly protected such as gender and ethnicity (bias variables), but may be indirectly related to the protected bias variables, such as a possible underlying correlation between ethnicity and bankruptcy. Claims 7-8 and 17-18 are rejected under 35 U.S.C. §103 as being unpatentable over Hacmon in view of Elovici, and further in view of Chalamalasetti et al. (US 2022/0121885 A1), hereinafter “Chalamalasetti”, and Duplessis et al. (US 2022/0245517 A1), hereinafter “Duplessis”. With respect to Claims 7 and 17: Hacmon in view of Elovici teaches: The computer implemented system of claim 1 and the computer implemented method of claim 11 respectively, wherein the computer processor is further configured to: ([0058] Hacmon teaches the sensitivity test that determines a bias output value. [0060] Hacmon teaches the likelihood ratio positive test that determines an output (i.e., materiality output value) based on the bias output value from the sensitivity test.) ([0058] Hacmon teaches the sensitivity test that determines a bias output value. [0060] Hacmon teaches the likelihood ratio positive test that determines an output (i.e., materiality output value) based on the bias output value from the sensitivity test.) Hacmon in view of Elovici does not explicitly teach: generate one or more backup snapshot instances of the trained machine learning, triggered by the estimated bias output value and the materiality output value being below a threshold data value; wherein the trained machine learning model is a reinforcement learning model. However, Chalamalasetti teaches in the same field of endeavor: the ([0042] Chalamalasetti teaches “determining whether the level of bias detected in the ML model 104 – as represented by one or more of the aforementioned statistical metrics -- is below a threshold level of bias.” Chalamalasetti clarifies that the statistical metrics may include [0039] average bias in the ML model (i.e., estimated bias output value) and [0041] standard deviation of bias. [0044] Chalamalasetti teaches that this below-threshold determination triggers a next step in “block 418”.) wherein the trained machine learning model is a reinforcement learning model. ([0018] Chalamalasetti teaches that the ML model may employ a reinforcement learning algorithm.) It would have been obvious for one of ordinary skill in the art prior to the effective filing date of the claimed invention to have modified Hacmon in view of Elovici’s teachings by comparing both the bias output value and a statistical value related to the materiality or confidence of the bias value to a threshold value for a reinforcement learning ML model as taught by Chalamalasetti. One would have been motivated to make this modification in order to further ensure that the bias detection is statistically significant or material enough to give further attention. This allows for increased accuracy of predictions in allowing for non-significant bias variables to remain in the data and provide a wider range of data for model training and testing. One would have been motivated to use a reinforcement learning model in order to apply the bias-testing process to more varied types of models such as an adaptive model trained on feedback data from the environment. Hacmon in view of Elovici and Chalamalasetti does not explicitly teach: periodically determine and record time-stamped versions of the estimated bias output value and the materiality output value for the trained machine learning model; and generate one or more backup snapshot instances of the trained machine learning, triggered by the estimated bias output value and the materiality output value being below a threshold data value; However, Duplessis teaches in the same field of endeavor: periodically determine and record time-stamped versions of the ([0013, 0084] Duplessis teaches tracking (i.e., recording time-stamped versions) model output values over a period of time for online monitoring (i.e., periodic determination of output values).) generate one or more backup snapshot instances of the trained machine learning,([0092] Duplessis teaches “certain models are stored as “snapshots” of themselves such that a model can be rolled back to an earlier trained state.”) It would have been obvious for one of ordinary skill in the art prior to the effective filing date of the claimed invention to have modified Hacmon in view of Elovici and Chalamalasetti’s teachings by testing and generating snapshots of a reinforcement learning model as taught by Duplessis. One would have been motivated to make this modification in order to provide appropriate backup measures such as when the target model presents an issue and must be replaced by a backup snapshot to maintain constant production [Duplessis 0113]. With respect to Claims 8 and 18: Hacmon in view of Elovici teaches: The computer implemented system of claim 1 and the computer implemented method of claim 11 respectively, wherein the computer processor is further configured to: ([0058] Hacmon teaches the sensitivity test that determines a bias output value. [0060] Hacmon teaches the likelihood ratio positive test that determines an output (i.e., materiality output value) based on the bias output value from the sensitivity test.) ([0058] Hacmon teaches the sensitivity test that determines a bias output value. [0060] Hacmon teaches the likelihood ratio positive test that determines an output (i.e., materiality output value) based on the bias output value from the sensitivity test.) Hacmon in view of Elovici does not explicitly teach: periodically determine and record time-stamped versions of the estimated bias output value and the materiality output value for the trained machine learning model; and generate one or more backup snapshot instances of the trained machine learning, triggered by the estimated bias output value and the materiality output value being below a threshold data value; wherein the trained machine learning model is a machine learning model being trained using new real-time data. However, Chalamalasetti teaches in the same field of endeavor: ([0042] Chalamalasetti teaches “determining whether the level of bias detected in the ML model 104 – as represented by one or more of the aforementioned statistical metrics -- is below a threshold level of bias.” Chalamalasetti clarifies that the statistical metrics may include [0039] average bias in the ML model (i.e., estimated bias output value) and [0041] standard deviation of bias. [0044] Chalamalasetti teaches that this below-threshold determination triggers a next step in “block 418”.) It would have been obvious for one of ordinary skill in the art prior to the effective filing date of the claimed invention to have modified Hacmon in view of Elovici’s teachings by comparing both the bias output value and a statistical value related to the materiality or confidence of the bias value to a threshold value as taught by Chalamalasetti. One would have been motivated to make this modification in order to further ensure that the bias detection is statistically significant or material enough to give further attention. This allows for increased accuracy of predictions in allowing for non-significant bias variables to remain in the data and provide a wider range of data for model training and testing. Hacmon in view of Elovici and Chalamalasetti does not explicitly teach: periodically determine and record time-stamped versions of the estimated bias output value and the materiality output value for the trained machine learning model; and generate one or more backup snapshot instances of the trained machine learning, triggered by the estimated bias output value and the materiality output value being below a threshold data value; wherein the trained machine learning model is a machine learning model being trained using new real-time data. However, Duplessis teaches in the same field of endeavor: periodically determine and record time-stamped versions of the ([0013, 0084] Duplessis teaches tracking (i.e., recording time-stamped versions) model output values over a period of time for online monitoring (i.e., periodic determination of output values).) generate one or more backup snapshot instances of the trained machine learning,([0092] Duplessis teaches “certain models are stored as “snapshots” of themselves such that a model can be rolled back to an earlier trained state.”) wherein the trained machine learning model is a machine learning model being trained using new real-time data. ([0008] Duplessis explains how the model may be trained in real-time in online learning. [0020, 0092] Duplessis explains how models may be stored as snapshots such that they can be rolled back to an earlier state in the case of a degraded current model.) It would have been obvious for one of ordinary skill in the art prior to the effective filing date of the claimed invention to have modified Hacmon in view of Elovici’s teachings by taking snapshots of models trained using real-time data as taught by Duplessis. One would have been motivated to make this modification in order to account for unintended bias that may appear during real-time training, allowing the model to roll back to a version prior to being trained on the dataset that had introduced bias to the model. Claims 9-10 and 19 are rejected under 35 U.S.C. §103 as being unpatentable over Hacmon in view of Elovici, Chalamalasetti, Wang, and further in view of Stillman et al. (US 2022/0191208 A1), hereinafter “Stillman”. With respect to Claims 9 and 19: Hacmon in view of Elovici, Chalamalasetti, and Wang teaches: The computer implemented system of claim 2 and the computer implemented method of claim 12 respectively, wherein the model pipeline control data process is configured to ([0032, 0106] Elovici explains that “[t]he system may be implemented over a Fronted Management Server [i.e., model pipeline control data process]” that can provide API access to perform one or more system operations.) Hacmon in view of Elovici, Chalamalasetti, and Wang does not explicitly teach: wherein the model pipeline control data process is configured to selectively control an exposed application programming interface (API). However, Stillman teaches in the same field of endeavor: wherein the model pipeline control data process is configured to selectively control an exposed application programming interface (API). ([0021] Stillman states “Businesses… may expose their data and receive data through APIs.” and “The controller 118 also controls what search engine APIs 122 are invoked.”) It would have been obvious for one of ordinary skill in the art prior to the effective filing date of the claimed invention to have modified Hacmon in view of Elovici, Chalamalasetti, and Wang’s teachings by selectively controlling exposed APIs. as taught by Stillman. One would have been motivated to make this modification in order to ensure that different ML models work upon the same convergent data in order to maintain accurate predictive capabilities (Stillman [0039]). With respect to Claim 10: Hacmon in view of Elovici, Chalamalasetti, Wang, and Stillman teaches: The computer implemented system of claim 9, wherein the model pipeline control data process ceases operation of the trained machine learning model by disconnecting the trained machine learning model from the exposed application programming interface (API). ([0055] Stillman describes how the controller (i.e., model pipeline control data process) may “terminat[e] an existing cross correlation between… the machine learning model… and one or more APIs.”) Conclusion The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Castiglione et al. (US 2022/0114399 A1) teaches fairness testing by employing an auxiliary ML model to compare against the gradient of the target ML model. Lee et al. (US 11,593,648 B2) teaches bias detection through comparing performance metrics against baseline metrics of the feature group. Lohia et al. (US 2021/0158102 A1) teaches a first dataset that contains protected attributes and a second dataset wherein the protected attributes have been removed. Mishraky et al. (US 2022/0351068 A1) discloses detecting latent bias in ML models, wherein identifying information is received, analyzed for a bias probability, fed into a ML model, and the output of the model is determined to be biased based on the probability. Mueck et al. (US 2024/0273411 A1) teaches retraining or interrupting the AI/ML system/models in the case that bias is detected. Any inquiry concerning this communication or earlier communications from the examiner should be directed to BRIAN D. BUI whose telephone number is (571)270-0463. The examiner can normally be reached Monday - Friday 8:00am - 5:00pm. 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, ABDULLAH AL KAWSAR can be reached at (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 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. /BRIAN D. BUI/Examiner, Art Unit 2127 /ABDULLAH AL KAWSAR/Supervisory Patent Examiner, Art Unit 2127
Read full office action

Prosecution Timeline

Jun 17, 2024
Application Filed
Oct 23, 2026
Non-Final Rejection mailed — §101, §103 (current)

Strategy Recommendation AI-generated — please review before filing

Get a prosecution strategy drawn from examiner precedents, rejection analysis, and claim mapping.
Typically takes 5-10 seconds — AI-generated, attorney review required before filing

Prosecution Projections

1-2
Expected OA Rounds
Grant Probability
Low
PTA Risk
Based on 0 resolved cases by this examiner. Grant probability derived from career allowance rate.

Sign in with your work email

Enter your email to receive a magic link. No password needed.

Personal email addresses (Gmail, Yahoo, etc.) are not accepted.

Free tier: 3 strategy analyses per month