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 .
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR
1.17(e), was filed in this application after final rejection. Since this application is eligible for continued
examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the
finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's
submission filed on 01/15/2026 has been entered.
Response to Amendment
The amendments filed 01/15/2026 have been entered. Claims 1-4, 6-8 remain pending in the application.
Applicant’s amendments and arguments, with respect to claim rejections of claims 1-4, 6-8 under 35 U.S.C 101 filed 08/14/2025 have been considered and are not persuasive. Therefore, the previous rejections as set forth in the previous office action has been maintained.
Applicant argues that the pending claims are not directed to an abstract idea and that the previous § 101 rejection improperly characterized the claims at too high a level of generality. Applicant contends that the claims should be reviewed as a whole under Step 2A, Prong Two, and that the examiner should not oversimplify the claimed invention as merely performing calculations or generating a second model.
Applicant further argues that the claimed limitations cannot practically be performed in the human mind, even with pen and paper, because the claims require generating a second model by performing machine learning. Applicant asserts that under recent USPTO guidance, the mental process category should not be expanded based only on theoretical possibility, and that a claim recites a mental process only when the limitations can practically be performed in the human mind.
Applicant also argues that the claims integrate any alleged abstract idea into a practical application because the claims allegedly improve a learned model by correcting unintended bias, such as gender or age bias. Applicant points to the specification as describing that conventional learned models do not sufficiently consider bias, and that the claimed process determines bias using a threshold and a ratio between first and second input values, substitutes a second input value to obtain a second prediction result, receives feedback on the second prediction result, and generates a second model by machine learning to correct the bias.
Applicant further asserts that the claimed invention is directed to an improvement in computer technology or an existing technological process because the claimed features allegedly improve fairness of a learned model and are not merely directed to a result or effect. Applicant relies on the claimed machine-learning model generation, feedback, and bias correction process as integrating the alleged abstract idea into a practical application under Step 2A, Prong Two.
Finally, under Step 2B, Applicant argues that the claims recite significantly more than well-understood, routine, and conventional activity because the claimed ordered combination allegedly provides a specific way to detect and correct bias in a learned model using a ratio, threshold, substituted input value, second prediction result, feedback, and second model generation.
The examiner respectfully disagrees. Applicant’s arguments have been considered but are not persuasive. The pending claims are not rejected merely because they involve a learned model or because they mention “fairness.” Rather, the rejection identifies that the claimed bias determination is based on comparing values, thresholds, categories, and ratios derived from prediction-result statistics. For example, the claims recite determining whether a prediction result satisfies a predetermined condition based on a threshold associated with first and second input values, a ratio between first and second parameter values, and whether the ratio is less than the threshold. These limitations amount to mathematical calculations and evaluations of information of value, which can be performed mentally or with pen and paper, and therefore fall within the abstract-idea groupings of mathematical concepts and mental processes.
Applicant’s argument that the claims improve a learned model or correct bias does not show integration into a practical application. The claims do not recite a specific improvement to a computer, processor, memory, network, model architecture, training algorithm, or machine-learning operation. Instead, the claims use a generic prediction system and first model as tools to obtain prediction results, apply the abstract bias/ratio/threshold analysis, substitute an input value, obtain a second prediction result, receive feedback, and generate a second model. These steps amount to applying the abstract idea using generic computer and machine-learning components, rather than improving the functioning of the computer or any other technology.
Nor do the additional limitations amount to significantly more under Step 2B. Receiving requests/results/feedback, preserving prediction results, and providing prediction results to a business system correspond to routine data gathering, storing, receiving, and transmitting information back and forth between the business system and the prediction system as identified in MPEP 2106.05(d) and MPEP 2106.05(g). Generating a second model from learning data is recited at a high level and merely applies a generic machine-learning model to the abstract bias-evaluation process. Considered individually and as an ordered combination, the claims do not add an inventive concept beyond evaluating prediction statistics to identify bias and using generic ML processing to apply that evaluation. Accordingly, the 101 rejections is maintained.
Applicant’s amendments and arguments, with respect to claim rejections of claims 1-4, 6-8 under 35 U.S.C 103 filed 08/14/2025 have been considered and are persuasive. However, upon further consideration, new ground(s) of rejections have been raised (See Below.)
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 1-4, 6-8 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an
abstract idea without significantly more.
Regarding claim 1,
Step 1:
Claim 1 recites a system, one of the four statutory categories of patentable subject matter.
Step 2A, Prong I:
Claim 1 further recites the limitations of:
“determine whether the first prediction result ... satisfies a predetermined condition based on” This limitation recites a mental process. A person can mentally determine if a prediction result satisfy a predetermined condition such as by comparing the prediction value to a threshold/criterion. Such determination can be carried out as a mental process via a human’s mind or manually performed via pen and paper.
“a predetermined threshold associated with the first input value and a second input value, which is a comparison input value of the first input value” This limitation recites a mental process as well as a mathematical concept. A person can mentally or manually determine a threshold and further compare the input values with the threshold or compare between two input values. Such comparison between values is a mathematical concept as well as a mental process that can be carried out via a human’s mind or manually performed via pen and paper.
“a ratio between a first parameter value associated with the first input value and a second parameter value associated with the second input value determined from prediction result statistics information associated with the first input value and the second input value” This limitation recites a mental process as well as a mathematical concept. A person can mentally or manually determine a ratio between statistic values. Such calculation of ratio between values is a mathematical concept as well as a mental process that can be carried out via a human’s mind or manually performed via pen and paper.
“wherein the first input value and the second input value both belong to a same category” This limitation recites a mental process. A person can mentally determine first and second input value that belong to the same category. For example, a person can mentally determined two separate set of values of male and female, wherein both of them belong to the same category.
“wherein the category is one of a plurality of categories including gender, nationality, locality, academic background, race, age, number of work years, and income, and” This limitation recites a mental process. A person can mentally determine any category for prediction such as gender, nationality, locality, academic background, race, age, number of work years, and income. A human’s mind is capable of determine these categories and predicting values relating to these categories. For example, a person can mentally predict that how many males would like a specific car design and how many females would like the same car design.
“wherein the predetermined condition is satisfied indicating that a bias exists, in a state where the ratio is less than the predetermined threshold” This limitation recites a mental process as well as a mathematical concept. A person can mentally determine a threshold of value and ratio between as indicated above, wherein these determination further constitute a mathematical concept that can be performed within a human’s mind. Furthermore, the person can mentally compare the ratio value with the threshold value to determine a satisfied condition, wherein the comparison between values also constitute a mathematical concept, as well as a mental process.
Step 2A, Prong II:
Claim 1 recites the following additional elements:
“at least one memory storing instructions; and at least one processor executing the instructions causing the system to” These additional elements are a high-level recitation of generic computer components used as a tool, and does not provide integration into a practical application.
“receive, from the business system, a request for a first prediction result to be obtained from the first model with an input including a first input value” This additional element recites an additional element of an insignificant extra-solution activity as identified in MPEP 2106.05(g) of mere data gathering, and does not provide integration into a practical application.
“... the first prediction result obtained from the first model with the input including the first input value ...” This additional element recites an additional element of a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), and does not provide integration into a practical application. The limitation recites the application of black-box machine learning model prediction. The claim recites the first model obtain the first input and provide the first prediction result, without providing the improvement over the prediction algorithm or unconventional machine learning prediction practice or improvement to a computer element. The claim simply recites using a model to perform prediction over an input.
“request a second prediction result from the first model included in the prediction system by substituting the first input value included in the input with the second input value, in a state where the first prediction result satisfies the predetermined condition” This additional element recites an additional element of a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), and does not provide integration into a practical application. The limitation recites the application of black-box machine learning model prediction. The claim recites evaluating the first prediction, then substitute the first input with the second input, and provide the second prediction result via the model. Such method is a conventional black-box machine learning practice where initial input is replaced with new input based on evaluation of the result obtained from the first input. Furthermore, the evaluation between prediction values is determined to be a mental process and mathematical concept as recited in Step 2A prong I above. Thus, the limitation does not demonstrate an improvement based on an innovative prediction algorithm or unconventional machine learning prediction practice or improvement to a computer element. The limitation simply recites using a model to continuously perform prediction without provide integration into a practical application
“preserve the second prediction result obtained from the prediction system” This additional element recites an additional element of an insignificant extra-solution activity as identified in MPEP 2106.05(g) of mere data gathering, and does not provide integration into a practical application.
“provide the second prediction result to the business system” This additional element recites an additional element of an insignificant extra-solution activity as identified in MPEP 2106.05(g) of mere data gathering, and does not provide integration into a practical application.
“receive feedback on the second prediction result from the business system” This additional element recites an additional element of an insignificant extra-solution activity as identified in MPEP 2106.05(g) of mere data gathering, and does not provide integration into a practical application.
“generate a second model by performing machine learning using first learning data formed by the input including the first input value and the second prediction result based on the received feedback” This additional element recites an additional element of a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), and does not provide integration into a practical application. The limitation recites the application of black-box machine learning model prediction. The limitation recites obtaining the second model by using first learning data from the fist input, the second prediction result, and feedbacks. Overall, the limitation recites conventional practice of a prediction machine learning model, in which the model continually learn from feedbacks over its prediction and its original input to update the model itself to obtain a newly improved model. Such iterative learning to obtain the improved model is a conventional machine learning black-box application. The limitation does not recite a specific configuration to configure the new model based on the acquired data and feedbacks or unconventional method of machine learning to generate a new improved model other than mentioning that a new model is obtained based on performing prediction and obtain feedbacks, which in view of the abstract idea of evaluating prediction values as recited in Step 2A Prong I above, does not constitute an improvement toward machine learning model practice, and does not provide integration into a practical application.
Step 2B:
When considered individually or in combination, the additional limitations and elements of claim 1 does not amount to significantly more than the judicial exception for the same reasons discussed above as to why the additional limitations do not integrate the abstract idea into a practical application. The additional elements of outlined in Step 2A performing functions as designed simply accomplishes execution of the abstract ideas.
The additional element “at least one memory storing instructions; and at least one processor executing the instructions causing the system to” is a high-level recitation of generic computer components used as a tool, and does not amount to significantly more than the judicial exception for the same reasons discussed above as to why the additional limitations do not integrate the abstract idea into a practical application.
The additional element “receive, from the business system, a request for a first prediction result to be obtained from the first model with an input including a first input value” further recites an additional element of a well-understood, routine, conventional activity as identified in MPEP 2106.05(d) of receiving or transmitting data over a network, and does not amount to significantly more than the judicial exception for the same reasons discussed above.
The additional element “... the first prediction result obtained from the first model with the input including the first input value ...” recites an additional element of a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), and does not amount to significantly more than the judicial exception for the same reasons discussed above.
The additional element “preserve the second prediction result obtained from the prediction system” further recites an additional element of a well-understood, routine, conventional activity as identified in MPEP 2106.05(d) of storing and retrieving information, and does not amount to significantly more than the judicial exception for the same reasons discussed above.
The additional element “provide the second prediction result to the business system” further recites an additional element of a well-understood, routine, conventional activity as identified in MPEP 2106.05(d) of receiving or transmitting data over a network, and does not amount to significantly more than the judicial exception for the same reasons discussed above.
The additional element “receive feedback on the second prediction result from the business system” This additional element further recites an additional element of a well-understood, routine, conventional activity as identified in MPEP 2106.05(d) of receiving or transmitting data over a network, and does not amount to significantly more than the judicial exception for the same reasons discussed above.
The additional element “request a second prediction result from the first model included in the prediction system by substituting the first input value included in the input with the second input value, in a state where the first prediction result satisfies the predetermined condition” recites an additional element of a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), and does not amount to significantly more than the judicial exception for the same reasons discussed above.
The additional element “generate a second model by performing machine learning using first learning data formed by the input including the first input value and the second prediction result based on the received feedback” recites an additional element of a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), and does not amount to significantly more than the judicial exception for the same reasons discussed above.
In conclusions from above for the elements considered as a mental process, elements reciting high-level recitation of generic computer components used as a tool, and elements reciting a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), elements reciting a well-understood, routine, conventional activity as identified in MPEP 2106.05(d), and elements reciting an insignificant extra-solution activity as identified in MPEP 2106.05(g) are carried over and do not provide significantly more than the abstract idea. Looking at the limitations in combination and the claims as a whole does not change this conclusion and the claim is ineligible.
Therefore, additional limitations of claim 1 do not amount to significantly more than the judicial exception.
Thus, claim 1 recites abstract ideas with additional elements rendered at a high level of generality resulting in claims that do not integrate the abstract idea into a practical application or amount to significantly more than the judicial exception.
Therefore, claim 1 is not patent eligible.
Regarding claim 2 depends on claim 1, thus the rejection of claim 1 is incorporated.
“The fairness management system according to claim 1, wherein the at least one processor executes the stored instructions further ...” This element is a high-level recitation of generic computer components used as a tool, and does not amount to significantly more than the judicial exception for the same reasons discussed above as to why the additional limitations do not integrate the abstract idea into a practical application.
“... causing the system to determine whether a prediction result obtained using a verification input by the second model satisfies the predetermined condition” This additional element recites an additional element of a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), and does not provide integration into a practical application or significantly more than the judicial exception. The limitation recites black box application of machine learning prediction, in which a model further perform prediction, evaluate the prediction to evaluate the model. Such practice is a conventional machine learning black-box application.
Thus, claim 2 recites additional elements rendered at a high level of generality resulting in claims that do not integrate the abstract idea into a practical application or amount to significantly more than the judicial exception.
Regarding claim 3 depends on claim 1, thus the rejection of claim 1 is incorporated.
“The fairness management system according to claim 2, wherein the at least one processor executes the stored instructions further ...” This element is a high-level recitation of generic computer components used as a tool, and does not amount to significantly more than the judicial exception for the same reasons discussed above as to why the additional limitations do not integrate the abstract idea into a practical application.
“... causing the system to replace the first model with the second model” This additional element recites an additional element of a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), and does not provide integration into a practical application or significantly more than the judicial exception. The limitation recites black box application of machine learning model update with better model based on its performance evaluation without providing significantly more such as machine learning algorithm to perform model substitution or unconventional model substitution practice. Instead, the limitation simply recites the replacement of a model with a better model, which is a black-box application of machine learning practice.
“... the prediction result obtained using the verification input by the second model is determined to not satisfy the predetermined condition” This element recites a mental process. A person can mentally or manually evaluate the prediction result from using a machine learning model by comparing its prediction value to a predetermined condition and determine if it satisfies. Such evaluation is a mental process.
Thus, claim 3 recites abstract ideas rendered at a high level of generality resulting in claims that do not integrate the abstract idea into a practical application or amount to significantly more than the judicial exception.
Regarding claim 4 depends on claim 1, thus the rejection of claim 1 is incorporated.
“The fairness management system according to claim 1, wherein: the at least one processor executes the stored instructions further”, “causing the system to ...” These elements are a high-level recitation of generic computer components used as a tool, and does not amount to significantly more than the judicial exception for the same reasons discussed above as to why the additional limitations do not integrate the abstract idea into a practical application.
“... generate second learning data in which at least some values of the input overlap based on the first learning data formed by the input including the first input value and the second prediction result, ...” This element recites a mental process. A person can mentally or manually perform value prediction and generate a different learning data with input values overlap with another learning data. Such data manipulation is capable to be performed within a human’s mind.
“... the second model is generated by performing machine learning using the first learning data formed by the input including the first input value and the second prediction result in addition to the second learning data” This additional element recites an additional element of a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), and does not provide integration into a practical application or significantly more than the judicial exception. The limitation recites the application of black-box machine learning model prediction. The limitation recites obtaining the second model by using first learning data from the fist input, second learning data, and the second prediction result. Overall, the limitation recites conventional practice of a prediction machine learning model, in which the model continually learn from feedbacks over its prediction and its original input to update the model itself to obtain a newly improved model. Such iterative learning to obtain the improved model is a conventional machine learning black-box application.
Thus, claim 4 recites abstract ideas with additional elements rendered at a high level of generality resulting in claims that do not integrate the abstract idea into a practical application or amount to significantly more than the judicial exception.
Regarding claim 6 depends on claim 1, thus the rejection of claim 1 is incorporated.
“... a first ratio of a number of times that a prediction result predicted using the input including the first input value by the first model is a predetermined prediction result to a total prediction number predicted using the input including the first input value by the first model is set as the first parameter value” This element recites a mental process as well as a mathematical concept. A person can mentally or manually determine a ratio between statistic values. Such calculation of ratio between values is a mathematical concept as well as a mental process that can be carried out via a human’s mind or manually performed via pen and paper.
“a second ratio of a number of times that a prediction result predicted using the input including the second input value by the first model is the predetermined prediction result to a total prediction number predicted using the input including the second input value by the first model is set as the second parameter value” This element recites a mental process as well as a mathematical concept. A person can mentally or manually determine a ratio between statistic values. Such calculation of ratio between values is a mathematical concept as well as a mental process that can be carried out via a human’s mind or manually performed via pen and paper.
Thus, claim 6 recites abstract ideas at a high level of generality resulting in claims that do not integrate the abstract idea into a practical application or amount to significantly more than the judicial exception.
Regarding claim 7, which recites a method, one of the four statutory categories of patentable subject matter. The applicant is directed to the rejection of claim 1 above because the claim recites similar limitations and processing steps.
Regarding claim 8, which recites a machine, one of the four statutory categories of patentable subject matter
Claim 8 recites the following additional elements:
“at least one memory storing instructions; and at least one processor executing the instructions causing the system to” These additional elements are a high-level recitation of generic computer components used as a tool, and does not provide integration into a practical application, and do not provide significantly more than the abstract idea.
The applicant is directed to the rejection of claim 1 above because the claim recites similar limitations and processing steps.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-4, 6-8 are rejected under 35 U.S.C. 103 as being unpatentable over Weider et.al (US 20200380399 A1), in view of Merai et.al (US 20190171897 A1), further in view of Feldman et.al (NPL: Certifying and removing disparate impact).
Regarding claim 1,
Weider teaches or at least suggest “A fairness management system communicable with a prediction system including a first model generated by machine learning and communicable with a business system, to manage the first model, the fairness management system comprising” (paragraph 2 “researchers conducting research or organizations that are in the business of collecting data are some of the entities that may provide datasets for training machine leaning models”, paragraph 27 “FIG. 1 illustrates system architecture 100 for detecting and visualizing bias in machine learning operations ... data relating to specific outcomes may be collected and provided by organizations that wish to use the outcomes to train models that predict more desirable outcomes”, and paragraph 30 “The bias detection system 120 may be provided as a service that can access and statistically examine a dataset to identify bias and/or imbalanced data... The user 170 may be a person(s) responsible for managing the ML training or any other user of a dataset in the dataset repository”. Weider discloses a bias detection system in communication with organization that wish to utilize machine learning model to perform prediction with more desirable outcomes, wherein the organization corresponds to or at least suggest the claimed business system, the machine learning model utilized by the organizations to perform prediction corresponds to a first model generated by machine learning in the prediction system as claimed, and the bias detection system corresponds to the fairness management system as the bias detection system detect bias in dataset to ensure fairness in dataset.)
Weider teaches “at least one memory storing instructions; and at least one processor executing the instructions causing the system to:” (paragraph 67 “The memory/storage 830 may include a main memory 832, a static memory 834, or other memory, and a storage unit 836, both accessible to the processors 810 such as via the bus 802. The storage unit 836 and memory 832, 834 store instructions 816 embodying any one or more of the functions” Weider discloses the function of the bias detection system is carried out via a memory storing instructions embodying any one or more of the functions, which is accessible to the processors to execute the instructions.)
Weider teaches or at least suggest “wherein the category is one of a plurality of categories including gender, nationality, locality, academic background, race, age, number of work years, and income, and” (paragraph 2 “Other institutions may utilize machine learning models to make hiring decisions, salary”, paragraph 27 “In another example, a dataset may be provided by a researcher conducting research on a population or a scientific subject. For example, health related data may be provided by researchers that conduct research in the medical and health fields and provide their findings in a dataset. Other types of data collection may be employed. For example, polling data may be collected and provided by pollsters ... For example, banks may collect data on loan defaults and circumstances”, paragraph 51 “For example, the common features may include gender, race, sexual orientation, and age”. Weider discloses organization/user may perform prediction with relate to hiring decisions, salary, health related data, polling data, circumstances, gender race, sexual orientation, and age, thereby teaches or at least suggests the categories for predictions including gender, nationality, locality, academic background, race, age, number of work years, and income as claimed. For instance, Weider’s disclosure of using machine-learning models to make hiring decisions teaches or at least suggests the claimed academic background category because hiring decisions commonly rely on applicant qualifications, such as education, degree, school, or academic credentials; Weider’s disclosure of polling data teaches or at least suggests the claimed locality category because polling data is commonly collected and analyzed based on geographic location, voting district, city, county, state, or region; Weider’s disclosure of health-related data teaches or at least suggests the claimed nationality category because health and medical research datasets may include nationality as population information for analyzing health outcomes; and Weider’s disclosure of banks collecting data on loan defaults and circumstances teaches or at least suggests the claimed number of work years category because loan/default analysis commonly considers borrower circumstances such as employment history, which may be represented by years worked.)
Weider teaches or at least suggest “provide the second prediction result to the business system” (paragraph 27 “organizations that wish to use the outcomes to train models that predict more desirable outcomes.” Weider discloses the organization obtain desired prediction results, which teaches or at least suggests that the second prediction results is provided to the business system as claimed.)
While Weider teaches a machine learning model that perform prediction and the organization that corresponds to the business system as recited above, Weider does not teach the first/second prediction sequence workflow as in the limitation “receive, from the business system, a request for a first prediction result to be obtained from the first model with an input including a first input value” However, Merai teaches or at least suggests this sequenced prediction workflow at (paragraph 46 “The input device 105 operates to gather electronic data that is representative of a real-world object and/or condition. It will be understood that the electronic data gathered by the input device 105 of the real-world object and/or condition will vary depending on the acquisition settings to which the input device 105 is configured”, paragraph 66 “That is, the predictor module 111 is configured to receive a first input data that was acquired using the data gathering device 105. The predictor module 111 is further configured to process the first input data to generate a first prediction from the input data and a corresponding confidence score of this first prediction”.Merai discloses receive a first input data from input device and generate a first prediction using the predictor module and first parameter, wherein the input device can gather data representative of a real-world object and/or condition, which teaches or at least suggests the first prediction result to be obtained from the first model with a first input, as claimed. In view of the combination with Weider’s teaching, the input device may correspond or relate to a business organization that attempt to perform machine learning prediction, thereby teaches or at least suggests the receiving the prediction request from the business system, as claimed.
Weider does not teach “determine whether the first prediction result obtained from the first model with the input including the first input value satisfies a predetermined condition based on:” However, Merai teaches or at least suggests this (paragraph 52 “A prediction of presence of an object and/or classification of the detected object can be determined as being successful if the confidence score generated by the predictor module 111 exceeds the predetermined confidence threshold”, paragraph 66 “That is, the predictor module 111 is configured to receive a first input data that was acquired using the data gathering device 105. The predictor module 111 is further configured to process the first input data to generate a first prediction from the input data and a corresponding confidence score of this first prediction.” Merai discloses generating a first prediction and further generate a corresponding confidence score of this first prediction, which is used to compare to a predetermined threshold to indicate if the prediction is success or not, thereby teaches or at least suggests the first prediction result satisfies a predetermined condition, as claimed.)
Weider does not teach “a predetermined threshold associated with the first input value and a second input value, which is a comparison input value of the first input value”. However, Merai teaches or at least suggests this (paragraph 52 “A prediction of presence of an object and/or classification of the detected object can be determined as being successful if the confidence score generated by the predictor module 111 exceeds the predetermined confidence threshold”, paragraph 73 “The action generation module 127 is further configured to calculate a difference between the confidence score of the first prediction (ft) and the confidence score of the second prediction”, paragraph 110 “At step 210, second input data acquired by the data gathering device is received. The second input data is acquired after the at least one action has been applied to the data gathering device ... The second input data is received by the predictor module”. Merai discloses a predetermined threshold in which the first prediction of the first input value is compared with to determine if it succeed in predicting, which corresponds to the predetermined threshold associated with the first input value, as claimed. Merai further discloses the second input data is acquired after the first prediction, wherein the second prediction is performed to calculate the difference between the first and second predictions to evaluate the accuracy of the predictions, thereby teaches or at least suggests the second input value as the comparison input value of the first input value, as claimed.)
Weider does not teach “wherein the first input value and the second input value both belong to a same category” However, Merai teaches or at least suggests this (paragraph 111 “... The captured real-world object is the same object captured in the first captured image received at step 202, but the image parameters will differ due to the application of the adjusted set of image capture settings.” Merai discloses that the first input and second input both belong to the same object but having different parameters setting, thereby teaches or at least suggests that the first and second input belong to the same category as claimed, because both inputs represent the same object, thus there is only one category of the object that both inputs can belong to.)
Weider does not teach “request a second prediction result from the first model included in the prediction system by substituting the first input value included in the input with the second input value, in a state where the first prediction result satisfies the predetermined condition” However, Merai teaches or at least suggests this (paragraph 52 “A prediction of presence of an object and/or classification of the detected object can be determined as being successful if the confidence score generated by the predictor module 111 exceeds the predetermined confidence threshold”, paragraph 64 “If the confidence ft is below the predetermined threshold θ, then assistant module 113 can be operated to tune input device 105 ... input device 105 will acquire a new input xt+1 and provide the same to predictor module 111. Upon receiving the new input xt+1, predictor module 111 will attempt to map the new input xt+1 to a best corresponding label”. Merai discloses after the predictor module perform the first prediction using the first input, the predictor module further acquires new input such as the second input and further perform the second prediction. The acquisition of the new input (second input) may be based on the comparison of the first prediction with predetermined confidence threshold to determine if the prediction is success or not to further attempt the second prediction, thereby teaches or at least suggests the claimed process of requesting a second prediction from the first model by substituting the first input value with the second input value, in a state where the first prediction result satisfies the predetermined condition, as claimed. The evaluating of the first and second prediction with a threshold may be further considered in view of the teaching by Feldman below.)
Weider does not teach “preserve the second prediction result obtained from the prediction system” However, Merai teaches or at least suggests this (paragraph 111 “At step 212, a second prediction is generated based on the second input data and a second confidence score for the second prediction is generated. The second prediction can be generated by the predictor module 111. ...”. Merai discloses the second prediction is generated by the predictor module and is further compared with the threshold to determine a difference with the first prediction, thus suggesting that the predictor module retain the second prediction information, thereby teaches or at least suggests that the preservation of the second prediction result obtained from the prediction system, as claimed.)
Weider in view of Merai teaches or at least suggest “receive feedback on the second prediction result from the business system” (Merai at paragraph 60 “Within each iteration of controlling input device 105 to gather addition data in the form of input 107, the parameters of the assistant module 113 for generating an appropriate feedback signal 115 can be adjusted over time based on which parameters in the feedback signal 115 led to an improvement in the subsequent prediction of subsequently received input 107 from the input device”. Merai discloses feedback signal from the input device which help led to an improvement in the subsequent prediction of subsequently received input, thereby teaches or at least suggests the receiving feedback on the second prediction result as claimed, wherein the input device may correspond to the business system in view of the teaching combination with Weider below.
Weider does not teach “generate a second model by performing machine learning using first learning data formed by the input including the first input value and the second prediction result based on the received feedback” However, Merai teaches or at least suggests this (paragraph 76 “the predictor module 111 includes an acquisitions settings adjustment model ... the adjustment model 136 can be retrained on an ongoing basis based on additionally acquired input data and differences in confidence scores in predictions resulting from sets of adjustment action generated by the action generation module ... updating the acquisition settings adjustment model 136 by machine learning, such as reinforcement learning ... The updating of the acquisition settings adjustment model can also be based on the first input data and the second input data”, paragraph 77 “The ongoing and iterative updating of the acquisition settings adjustment model 136 of the action generation module 127 based on iterations of acquisition of data and the adjustment of the data gathering device to acquire further data has the aim of improving the efficacy of the manner in which the data gathering device is configured and/or adjusted to improve input quality”. Merai discloses the predictor module include an adjustment model to perform prediction and retraining/updating the model by machine learning based on additionally acquired input data (second input) along with feedback signals as disclosed above, and differences in confidence between the predictions. As discussed above, Merai’s second input data is used to generate the second prediction, thus the differences include the second prediction. Furthermore, the updating/retraining uses information including the first prediction from the first input data, thus Merai discloses an updated/retrained model using the machine learning prediction from the first input data and the second prediction result along with its feedback signal, thereby teaches or at least suggest the claimed process, wherein the updated/retrained model suggests the second model, the model perform the machine learning prediction using the first input value corresponds to the claimed performing machine learning using first learning data formed by the first input value, and the model’s second prediction with feedback signals corresponds to the claimed second prediction result based on the received feedback.
Before the effective filing date, it would have been obvious to one of ordinary skill in the art to combine the teaching of the organization that perform machine learning prediction with the bias detection system by Weider with the teaching of system and method for automatically improving gathering of data including a first and second prediction comparison by Merai. The motivation is referred to in Merai’s disclosure (paragraph 5 “Machine learning algorithms depend heavily on input quality; if the input does not provide enough features, the machine learning algorithm will likely fail its prediction”, paragraph 6 “In order to overcome this problem, human intervention is often necessary to adjust the data collection device to ensure that sufficient input quality is provided”, and paragraph 56 “Input device 105 is subsequently tuned and/or adjusted, preferably so that it can acquire new data and provide an improved input 107 to processor 101 for subsequent prediction. As can be appreciated, this process can be repeated until input 107 is of sufficient quality such that predictor 111 can generate an output 109 with success”, paragraph 77 “The ongoing and iterative updating of the acquisition settings adjustment model 136 of the action generation module 127 based on iterations of acquisition of data and the adjustment of the data gathering device to acquire further data has the aim of improving the efficacy of the manner in which the data gathering device is configured and/or adjusted to improve input quality. As described elsewhere, the efficacy of the machine learning process is heavily dependent on input quality of the input data. The updating of the acquisition settings adjustment model, such as through reinforcement learning, can lead to the data gathering device 105 being appropriately reconfigured/adjusted so that a successful prediction ... occurs with greater frequency.” Merai discloses the benefit of improving machine-learning prediction by improving the quality of the input data. Specifically, Merai explains that machine-learning algorithms depend heavily on input quality, and if the input does not provide enough features, the machine-learning algorithm may fail its prediction. Merai further teaches evaluating predictions based on threshold, adjusting the data gathering device to acquire new or improved input data, generating a subsequent prediction, and iteratively updating the adjustment model so that successful predictions occur with greater frequency. Thus, a person of ordinary skill in the art would have been motivated to apply Merai’s first/second input and subsequent prediction-comparison framework with a threshold to Weider’s bias detection system. Weider teaches using machine-learning predictions to detect bias, but does not expressly disclose the claimed numbered first input/first prediction and second input/second prediction process. Applying Merai’s subsequent predictions framework to evaluate and improve prediction’s input would improve Weider’s machine learning prediction and utilization of dataset by providing a known technique for obtaining a comparison prediction from adjusted or additional input data, thereby allowing Weider’s system to better evaluate the prediction result and improve the reliability of the bias-detection process.)
Weider/Merai does not teach the limitation “a ratio between a first parameter value associated with the first input value and a second parameter value associated with the second input value determined from prediction result statistics information associated with the first input value and the second input value” However, Feldman teaches or at least suggest the limitation (Page 1 “In U.S. law, unintentional bias is encoded via disparate impact, which occurs when a selection process has widely different outcomes for different groups”, Page 4-5 section 3 “We start by reinterpreting the “80% rule” in terms of more standard statistical measures of quality of a classifier ... X takes on two values: X = 0 for the “minority” class and X = 1 for the “default” class. For example, in most gender-discrimination scenarios the value 0 would be assigned to “female” and 1 to “male”. We will denote a successful binary classification outcome C (say, a hiring decision) by C =YES and a failure by C =NO. Finally, we will map the majority class to “positive” examples and the minority class to “negative” examples with respect to the classification outcome ... The 80%rule can then be quantified as:
c
/
(
a
+
c
)
d
/
(
b
+
d
)
≥
0.8
...”, and Page 7 section 4.1 “DI threshold of τ = 0.8” Feldman discloses that protected class X has two values, X=0 for the “minority” class and X=1 for the “default” class, and further explains that in a gender-discrimination scenario, 0 may be assigned to “female” and 1 to “male.” This teaches or suggests the claimed first input value and second input value because female and male are different values of the same input attribute/category, i.e., gender, included in the data used for classification. Feldman discloses that a successful classification outcome C, such as a hiring decision, is denoted as C=YES. This teaches or suggests the claimed prediction result because the classifier produces an outcome/decision, such as YES or NO, based on the input data. Feldman discloses c/(a+c) as the favorable outcome rate for X=0 and d/(b+d) as the favorable outcome rate for X=1. This teaches or suggests the claimed first parameter value and second parameter value because each value is a statistic indicating how often a favorable prediction result occurs for the respective input value. Feldman discloses the 80% rule of greater or equal to 0.8, wherein the
c
/
(
a
+
c
)
d
/
(
b
+
d
)
teaches or suggests the claimed ratio between the first parameter value and the second parameter value because the formula divides the favorable outcome rate for X=0 by the favorable outcome rate for X=1. Feldman’s 0.8 teaches or suggests the claimed predetermined threshold, and a ratio below 0.8 indicates disparate impact/bias, thereby teaching or suggesting the claimed condition indicating bias when the ratio is less than the predetermined threshold.
Merai does not teach the limitation “wherein the predetermined condition is satisfied indicating that a bias exists, in a state where the ratio is less than the predetermined threshold” However, Feldman teaches or at least suggest the limitation (Page 4-5 section 3 “We start by reinterpreting the “80% rule” in terms of more standard statistical measures of quality of a classifier ... X takes on two values: X = 0 for the “minority” class and X = 1 for the “default” class. For example, in most gender-discrimination scenarios the value 0 would be assigned to “female” and 1 to “male”. We will denote a successful binary classification outcome C (say, a hiring decision) by C =YES and a failure by C =NO. Finally, we will map the majority class to “positive” examples and the minority class to “negative” examples with respect to the classification outcome ... The 80%rule can then be quantified as:
c
/
(
a
+
c
)
d
/
(
b
+
d
)
≥
0.8
...” Feldman discloses the 80% rule of greater or equal to 0.8, wherein the
c
/
(
a
+
c
)
d
/
(
b
+
d
)
teaches or suggests the claimed ratio between the first parameter value and the second parameter value because the formula divides the favorable outcome rate for X=0 (female) by the favorable outcome rate for X=1 (male). Feldman’s 0.8 teaches or suggests the claimed predetermined threshold, and a ratio below 0.8 indicates disparate impact/bias, thereby teaching or suggesting the claimed condition indicating bias when the ratio is less than the predetermined threshold.
Before the effective filing date, it would have been obvious to one of ordinary skill in the art to combine the teaching of the organization that perform machine learning prediction with the bias detection system by Weider, and the teaching of system and method for automatically improving gathering of data including a first and second prediction comparison by Merai, with the teaching of determining disparate impact (bias) via a ratio and threshold by Feldman. The motivation is referred to in Feldman’s disclosure (page 2 section 1 “The disparate impact certification problem is to guarantee that, given D, any classification algorithm aiming to predict some C (which is potentially different from the given C) from Y would not have disparate impact. By certifying any outcomes C, and not the process by which they were reached, we follow legal precedent in making no judgment on the algorithm itself, and additionally ensure that potentially sensitive algorithms remain proprietary”, and page 3 section 1.1 “We show that our algorithm certifying lack of disparate impact on a data set is effective, such that with the three classifiers we used certified data sets don’t show disparate impact. We demonstrate the fairness / utility tradeoff for our partial repair procedures. Comparing to related work, we find that for any desired fairness value we can achieve a higher accuracy than other fairness procedures. This is likely due to our emphasis on changing the data to achieve fairness, thus allowing any strong classifier to be used for prediction. Our procedure for detecting disparate impact goes through an actual classification algorithm.” Feldman discloses the benefit of detecting/certifying disparate impact so that prediction outcomes do not produce biased results for different groups. Specifically, Feldman explains that the disparate-impact certification problem is to ensure that a classification algorithm’s predicted outcome does not have disparate impact, and further teaches detecting disparate impact through an actual classification algorithm using a ratio-and-threshold analysis. Thus, Feldman cures the deficiency of the Weider/Merai combination because, although Weider teaches bias detection and Merai teaches a first/second input and prediction-comparison framework, the combination does not expressly teach determining bias using a ratio between first and second prediction-result statistics and comparing that ratio to a predetermined threshold. Feldman provides this missing feature by teaching the 80% rule, where the ratio of favorable outcome rates for two groups is compared to the threshold 0.8, and a ratio below the threshold indicates disparate impact/bias. Therefore, a POSITA would have been motivated to apply Feldman’s known ratio-and-threshold disparate-impact test to the Weider/Merai system and method to quantitatively determine when the compared prediction results indicate bias, thereby improve the dataset used for prediction.)
Regarding claim 2 depends on claim 1, thus the rejection of claim 1 is incorporated.
Merai teaches or at least suggests the limitation “The fairness management system according to claim 1, wherein the at least one processor executes the stored instructions further causing the system to determine whether a prediction result obtained using a verification input by the second model satisfies the predetermined condition” (paragraph 79 “Continuing from above, after first input data (xt) and second input data (xt+1) has been acquired by data gathering device 105, a third input data (xt+m) acquired by the data gathering device 105 can be received by the predictor module 111. The third input data (xt+m) is processed by the predictor module 111 to generate a third prediction and a corresponding confidence score ... It will be further appreciated that acquisitions settings adjustment model 136 can be updated based on the difference between the first prediction applied to the first input data and the second prediction applied to the second input data.”, and paragraph 80 “If it is determined that the third prediction is of insufficient quality (ex: confidence score of the third prediction is below the predetermined threshold)” Merai discloses a third input acquire by the model after the predictor module with its adjustment model has been updated based on the first and second prediction, which teaches or at least suggest a verification input obtained for the second model, as claimed. The module further provide prediction using the third input and calculate a confidence score to compare the prediction with a threshold, thereby teaches or at least suggests the prediction result obtained using a verification input by the second model satisfies the predetermined condition, as claimed.)
Regarding claim 3 depends on claim 2, thus the rejection of claim 2 is incorporated.
Merai teaches or at least suggests the limitation “The fairness management system according to claim 2, wherein the at least one processor executes the stored instructions further causing the system to replace the first model with the second model when the prediction result obtained using the verification input by the second model is determined to not satisfy the predetermined condition” (paragraph 79 “Continuing from above, after first input data (xt) and second input data (xt+1) has been acquired by data gathering device 105, a third input data (xt+m) acquired by the data gathering device 105 can be received by the predictor module 111. The third input data (xt+m) is processed by the predictor module 111 to generate a third prediction and a corresponding confidence score ... It will be further appreciated that acquisitions settings adjustment model 136 can be updated based on the difference between the first prediction applied to the first input data and the second prediction applied to the second input data.”, and paragraph 80 “If it is determined that the third prediction is of insufficient quality (ex: confidence score of the third prediction is below the predetermined threshold)” Merai discloses that, after first input data and second input data have been acquired, third input data may be received by the predictor module and processed to generate a third prediction and corresponding confidence score. Merai further discloses that the acquisition settings adjustment model may be updated based on the difference between the first prediction applied to the first input data and the second prediction applied to the second input data, and that the third prediction may be determined to be of insufficient quality when the confidence score is below a predetermined threshold. Thus, Merai teaches or at least suggests using the third input data as a verification input and determining whether the prediction result obtained using the verification input satisfies the predetermined condition. A person of ordinary skill in the art would have understood that the updated/retrained model is used in place of the prior model for subsequent predictions, thereby teaching or at least suggesting replacing the first model with the second model when the verification prediction does not satisfy the predetermined condition
Regarding claim 4 depends on claim 1, thus the rejection of claim 1 is incorporated.
Merai teaches or at least suggests the limitation “The fairness management system according to claim 1, wherein: the at least one processor executes the stored instructions further causing the system to generate second learning data in which at least some values of the input overlap based on the first learning data formed by the input including the first input value and the second prediction result, and the second model is generated by performing machine learning using the first learning data formed by the input including the first input value and the second prediction result in addition to the second learning data” (paragraph 76 “the predictor module 111 includes an acquisitions settings adjustment model ... the adjustment model 136 can be retrained on an ongoing basis based on additionally acquired input data and differences in confidence scores in predictions resulting from sets of adjustment action generated by the action generation module ... updating the acquisition settings adjustment model 136 by machine learning, such as reinforcement learning ... The updating of the acquisition settings adjustment model can also be based on the first input data and the second input data”, paragraph 77 “The ongoing and iterative updating of the acquisition settings adjustment model 136 of the action generation module 127 based on iterations of acquisition of data and the adjustment of the data gathering device to acquire further data has the aim of improving the efficacy of the manner in which the data gathering device is configured and/or adjusted to improve input quality”, and paragraph 111 “In the context of image recognition, the second prediction is an image recognition and/or classification of an object of the scene captured in the second image. The captured real-world object is the same object captured in the first captured image received at step 202, but the image parameters will differ” Merai discloses that the adjustment model may be retrained/updated by machine learning based on first input data, second input data, and differences in confidence scores from the first and second predictions. The first input data teaches or suggests the claimed input including the first input value, and the second prediction/confidence-score information teaches or suggests the claimed second prediction result. A person of ordinary skill in the art would have understood that the combination of input data and prediction-result information used to retrain/update a machine-learning model constitutes learning data; therefore, Merai teaches or suggests the claimed first learning data formed by the first input value with its first prediction result, and the second learning data formed by the second input and the second prediction result. Merai further discloses that the second image data captures the same real-world object as the first image data. Thus, Merai teaches or suggests second learning data having at least some values overlapping with the first learning data, because the first and second input data share the same captured object/input context while differing in parameters. Accordingly, Merai teaches or suggests training/updating the second model using the first learning data in addition to the second learning data, as claimed.)
Regarding claim 6 depends on claim 1, thus the rejection of claim 1 is incorporated.
Feldman teaches or at least suggests the limitation “The fairness management system according to claim 1, wherein: a first ratio of a number of times that a prediction result predicted using the input including the first input value by the first model is a predetermined prediction result to a total prediction number predicted using the input including the first input value by the first model is set as the first parameter value” (Page 4-5 section 3 “We start by reinterpreting the “80% rule” in terms of more standard statistical measures of quality of a classifier ... X takes on two values: X = 0 for the “minority” class and X = 1 for the “default” class. For example, in most gender-discrimination scenarios the value 0 would be assigned to “female” and 1 to “male”. We will denote a successful binary classification outcome C (say, a hiring decision) by C =YES and a failure by C =NO. Finally, we will map the majority class to “positive” examples and the minority class to “negative” examples with respect to the classification outcome ... The 80%rule can then be quantified as:
c
/
(
a
+
c
)
d
/
(
b
+
d
)
≥
0.8
...” Feldman discloses that protected class X takes on two values, including X=0 for the minority class, and in a gender-discrimination scenario, X=0 may be assigned to “female.” Feldman further discloses a successful classification outcome C=YES, such as a hiring decision. In Feldman’s 80% rule formula, c/(a+c) represents the number/rate of successful outcomes C=YES for X=0 relative to the total outcomes for X=0. Thus, c/(a+c) teaches or suggests the claimed first ratio of the number of times a predetermined prediction result is predicted using the input including the first input value to the total prediction number using the input including the first input value, and therefore teaches or suggests the claimed first parameter value.)
Feldman teaches or at least suggests the limitation “a second ratio of a number of times that a prediction result predicted using the input including the second input value by the first model is the predetermined prediction result to a total prediction number predicted using the input including the second input value by the first model is set as the second parameter value” (Page 4-5 section 3 “We start by reinterpreting the “80% rule” in terms of more standard statistical measures of quality of a classifier ... X takes on two values: X = 0 for the “minority” class and X = 1 for the “default” class. For example, in most gender-discrimination scenarios the value 0 would be assigned to “female” and 1 to “male”. We will denote a successful binary classification outcome C (say, a hiring decision) by C =YES and a failure by C =NO. Finally, we will map the majority class to “positive” examples and the minority class to “negative” examples with respect to the classification outcome ... The 80%rule can then be quantified as:
c
/
(
a
+
c
)
d
/
(
b
+
d
)
≥
0.8
...” Feldman discloses X=1 for the default class, and in a gender-discrimination scenario, X=1 may be assigned to “male.” Feldman further discloses a successful classification outcome C=YES, such as a hiring decision. In Feldman’s 80% rule formula, d/(b+d) represents the number/rate of successful outcomes C=YES for X=1 relative to the total outcomes for X=1. Thus, d/(b+d) teaches or suggests the claimed second ratio of the number of times a predetermined prediction result is predicted using the input including the second input value to the total prediction number using the input including the second input value, and therefore teaches or suggests the claimed second parameter value.)
Regarding claim 7, the applicant is directed to the rejection of claim 1 above because the claim recites similar limitations and processing steps.
Regarding claim 8,
Weider teaches “A non-transitory storage medium storing a program executable by machine learning ...” (paragraph 7 “the instant application describes a non-transitory computer readable medium on which are stored instructions that when executed cause a programmable device” Weider discloses the embodiment of the bias detection system may be carried out by a non-transitory computer readable medium on which are stored instructions to execute the system and method.)
The applicant is directed to the rejection of claim 1 above because the claim recites similar limitations and processing steps.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to DUY TU DIEP whose telephone number is (703)756-1738. The examiner can normally be reached M-F 8-4:30.
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, Alexey Shmatov can be reached at (571) 270-3428. 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.
/DUY T DIEP/ Examiner, Art Unit 2123
/ALEXEY SHMATOV/ Supervisory Patent Examiner, Art Unit 2123
Prosecution Insights