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 .
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and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114.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 7/30/2026 has been entered.
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 1-2, 4-11, 13-19, 21-23 are rejected under 35 U.S.C. 101
because the claimed invention is directed to an abstract idea without significantly
more.
When considering subject matter eligibility under 35 U.S.C. 101, it must be
determined whether the claim is directed to one of the four statutory categories of
invention, i.e., process, machine, manufacture, or composition of matter (Step 1). If the
claim does fall within one of the statutory categories, the second step in the analysis is to determine whether the claim is directed to a judicial exception (Step 2A). The Step 2A
analysis is broken into two prongs. In the first prong (Step 2A, Prong 1), it is determined
whether or not the claims recite a judicial exception (e.g., mathematical concepts,
mental processes, certain methods of organizing human activity). If it is determined in
Step 2A, Prong 1 that the claims recite a judicial exception, the analysis proceeds to the
second prong (Step 2A, Prong 2), where it is determined whether or not the claims
integrate the judicial exception into a practical application. If it is determined at step 2A,
Prong 2 that the claims do not integrate the judicial exception into a practical
application, the analysis proceeds to determining whether the claim is a patent-eligible
application of the exception (Step 2B). If an abstract idea is present in the claim, any
element or combination of elements in the claim must be sufficient to ensure that the
claim integrates the judicial exception into a practical application, or else amounts to
significantly more than the abstract idea itself. Applicant is advised to consult the 2019
PEG for more details of the analysis.
Step 1
According to the first part of the analysis, in the instant case, claims 1-2, 4, 21, 23, 5-11, 13-19, 22 are directed to a method, method and media of training ML models. Thus, each of the claims falls within one of the four statutory categories (i.e. process, machine, manufacture, or composition of matter). Step 2A,
Step 2A, Prong 1
Following the determination of whether or not the claims fall within one of the four
categories (Step 1), it must be determined if the claims recite a judicial exception (e.g.
mathematical concepts, mental processes, certain methods of organizing human
activity) (Step 2A, Prong 1). In this case, the claims are determined to recite a judicial
exception as explained below.
Regarding Claims 1, 5 and 13:
The claims recite a mental process. As set forth in MPEP 2106.04(a)(2)(III)(C), “Claims can recite a mental process even if they are claimed as being performed on a computer”. These are recited at a high level and disclosed as a human user performing these functions, simply using a computer as a tool-see spec, [0034-0048] etc. Fig. 1. etc. Thus, the claim recites abstract ideas.
Step 2A, Prong 2
Following the determination that the claims recite a judicial exception, it must be
determined if the claims recite additional elements that integrate the exception into a practical application of the exception (Step 2A, Prong 2). In this case, after considering all claim elements individually and as an ordered combination, it is determined that the claims do not include additional elements that integrate the exception into a practical application of the exception as explained below.
In Prong Two, a claim is evaluated as a whole to determine whether the recited judicial exception is integrated into a practical application of that exception. A claim is not “directed to” a judicial exception, and thus is patent eligible, if the claim as a whole integrates the recited judicial exception into a practical application of that exception. A claim that integrates a judicial exception into a practical application will apply, rely on, or use the judicial exception in a manner that imposes a meaningful limit on the judicial exception, such that the claim is more than a drafting effort designed to monopolize the judicial exception. MPEP 2106.04(d). The claims recite an abstract idea and further the claims as a whole does not integrate the recited judicial exception into a practical application of the exception. A claim that integrates a judicial exception into a practical application will apply, rely on, or use the judicial exception in a manner that imposes a meaningful limit on the judicial exception, such that the claim is more than a drafting effort designed to monopolize the judicial exception. MPEP 2106.04(d).
Regarding Claims 1, 5 and 13 these claims
This limitation recites using one or more neural networks as a tool to perform an
abstract idea, which is not indicative of integration into a practical application. MPEP 2106.05(f).)
This limitation is understood to be generic computer equipment and mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.0S(f))
Step 2B
Based on the determination in Step 2A of the analysis that the claims are
directed to a judicial exception, it must be determined if the claims contain any element
or combination of elements sufficient to ensure that the claim amounts to significantly
more than the judicial exception (Step 2B). In this case, after considering all claim
elements individually and as an ordered combination, it is determined that the claims do
not include additional elements that are sufficient to amount to significantly more than
the judicial exception for the same reasons given above in the Step 2A, Prong 2
analysis. Furthermore, each additional element identified above as being insignificant
extra-solution activity is also well-known, routine, conventional as described below.
Claims 1, 5 and 13: The claims do not include additional elements, alone or in combination, that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional elements amount to no more than generic computing components and field of use/technological environment which do not amount to significantly more than the abstract idea. The underlying concept merely receives information, analyzes it, and store the results of the analysis – this concept is not meaningfully different than concepts found by the courts to be abstract (see Electric Power Group, collecting information, analyzing it, and displaying certain results of the collection and analysis; see Cybersource, obtaining and comparing intangible data; see Digitech, organizing information through mathematical correlations; see Grams, diagnosing an abnormal condition by performing clinical tests and thinking about the results; see Cyberfone, using categories to organize store and transmit information; see Smartgene, comparing new and stored information and using rules to identify options). The claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception because the additional elements when considered both individually and as a combination do not amount to significantly more than the abstract idea. For example, claim 1 recites “obtain…”, “input…”, “input…”, “determine…”, “train…”., generate…”, etc. Claim 5 recite ““generating…”, “obtaining…”, “inputting…”, “inputting…”, “determining…”, “transmitting…”, etc. Claim 13 recites ““generating…”, “determining…”, “transmitting…”, etc. These elements are recited at a high level of generality and are well-understood, routine, and conventional activities in the computer art. Generic computers performing generic computer functions, without an inventive concept, do not amount to significantly more than the abstract idea. Looking at the elements as a combination does not add anything more than the elements analyzed individually. Therefore, these claims do not amount to significantly more than the abstract idea itself.
Step 2A/2B Prong 2 Dependent Claims
Regarding to claim 2, 6, 14
Claim 2, 6, 14 merely recite other additional elements that define generating bins or groups which performing generic functions that when looking at the elements as a combination does not add anything more than the elements analyzed individually. Therefore, these claims also do not amount to significantly more than the abstract idea itself. These claims are not patent eligible.
Regarding to claim 4
Claim 4 merely recite other additional elements that define determining residual scores which performing generic functions that when looking at the elements as a combination does not add anything more than the elements analyzed individually. Therefore, these claims also do not amount to significantly more than the abstract idea itself. These claims are not patent eligible.
Regarding to claim 7
Claim 7 merely recite other additional elements that define obtaining the set of entries which performing generic functions that when looking at the elements as a combination does not add anything more than the elements analyzed individually. Therefore, these claims also do not amount to significantly more than the abstract idea itself. These claims are not patent eligible.
Regarding to claim 8, 16
Claim 8, 16 merely recite other additional elements that define determining the type of data and sorting which performing generic functions that when looking at the elements as a combination does not add anything more than the elements analyzed individually. Therefore, these claims also do not amount to significantly more than the abstract idea itself. These claims are not patent eligible.
Regarding to claim 9-11, 17-19
Claim 9-11, 17-19 merely recite other additional elements that define determining error scores which performing generic functions that when looking at the elements as a combination does not add anything more than the elements analyzed individually. Therefore, these claims also do not amount to significantly more than the abstract idea itself. These claims are not patent eligible.
Regarding to claim 15
Claim 15 merely recite other additional elements that define determining group definitions which performing generic functions that when looking at the elements as a combination does not add anything more than the elements analyzed individually. Therefore, these claims also do not amount to significantly more than the abstract idea itself. These claims are not patent eligible.
Regarding to claim 21-22
Claim 21-22 merely recite other additional elements that define training the models and generating predictions which performing generic functions that when looking at the elements as a combination does not add anything more than the elements analyzed individually. Therefore, these claims also do not amount to significantly more than the abstract idea itself. These claims are not patent eligible.
Regarding to claim 23
Claim 23 merely recite other additional elements that define the category of data which performing generic functions that when looking at the elements as a combination does not add anything more than the elements analyzed individually. Therefore, these claims also do not amount to significantly more than the abstract idea itself. These claims are not patent eligible
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, 2, 5-7, 13-15, 21-23 are rejected under 35 U.S.C. 103 as being unpatentable over Lee et al. (“Methods and Systems for Detection and Isolation of Bias in Predictive Models”—U.S. Patent No. 11,593,648; hereinafter “Lee”) in view of Grossman US 12165017
Regarding claim 1
Regarding claim 1, Lee teaches a system for detecting bin-specific model bias and presenting of alternative predictions, the system comprising:
one or more processors; and one or more non-transitory computer-readable storage media storing instructions, which when executed by the one or more processors cause the one or more processors to (paragraph [0129], “The computing system 1000 executes program code that configures the processor 1102 to perform one or more of the operations described herein. The program code may be resident in the memory device 1008 or any suitable computer-readable medium and may be executed by the processor 1004 or any other suitable processor”; “memory device 1008” implies the use of tangible hardware, therefore a non-transitory computer-readable storage medium);
subsequent to a machine learning model being trained using a training dataset, generate a plurality of bins for a compare dataset; (paragraph [0066]-[0067], “A block 304, a computing device receives data associated with a trained predictive model. The data includes data output from the predictive model such as predictions associated with a class or a subclass. In some instances, the data includes the input data that was input to the predictive model to generate the output and training data used during a training phase of the predictive model… At block 308, the computing device identifies a feature group from the data”; paragraph [0054], “A feature group is made up of one or more features and represents a subclass of the class that includes members that each include characteristics that correspond to the one or more features”; feature groups can be considered analogous to parameters, and they may contain one or more features that intuitively correspond to bins of values);
obtain a training set of entries corresponding to a first bin of the plurality of bins from the training dataset and a compare set of entries, corresponding to the first bin, from the compare dataset (paragraph [0042], “In some instances, model test node 104 receives a seed from the training datasets 132… The model test node 104 uses the seed to procedurally generate any amount of training data that corresponds to the predictive model. The seed may be combined with a particular feature or feature group to generate training data that corresponds to that feature or feature group”; paragraph [0038], “For instance, the bias detection module 124 enumerates a set of features by identifying types of predictions that are generated, characteristics of a class (e.g., employees) that are the subject of the generated predictions, and/or statistically relevant aspects of the training data 208 or input data 212 such as data types, data structures, reoccurring data values, and the like”; input data is analogous to the compare dataset in this example; model test node generates (“obtains”) training data and bias detection module takes input data from feature sets);
input the training set of entries into the machine learning model to obtain a first set of residual values, (paragraph [0060], “The results of bias detection module 220 includes, but is not limited to, an identification of a bias associated with a feature group, a performance characteristic of the predictive model, an error, combinations thereof, an identification of a cause of the bias or error such as a characteristic of the predictive model 204, a portion of the training data 208, a portion of the input data 212, combinations thereof, or the like”; training data given to predictive model and error is output);
input the compare set of entries into the machine learning model to obtain a second set of residual values; (paragraph [0060], “The results of bias detection module 220 includes, but is not limited to, an identification of a bias associated with a feature group, a performance characteristic of the predictive model, an error, combinations thereof, an identification of a cause of the bias or error such as a characteristic of the predictive model 204, a portion of the training data 208, a portion of the input data 212, combinations thereof, or the like”; input data given to predictive model and error is output);
using the updated training dataset to train the alternative machine learning model associated with the first bin (paragraph [0065], “If the model correction includes a modification to the predictive model 204, the system 200 applies the model correction to the predictive model 204. If the model correction includes a modification to the training data 208, the system 200 applies the model correction to modify training data 208. Once the training data 208 is modified, the predictive model 204 is retraining using the modified training data 208… Alternatively, a new model correction may be developed and applied to the predictive model 204 and/or the training data 208”; modified predictive model is analogous to an alternative machine model, which may then be retrained on modified training data);
in response to the machine learning model generating a prediction from production data input matching the first bin, input the production data input into the alternative machine learning model (paragraph [0042], “In some instances, model test node 104 receives a seed from the training datasets 132… The model test node 104 uses the seed to procedurally generate any amount of training data that corresponds to the predictive model. The seed may be combined with a particular feature or feature group to generate training data that corresponds to that feature or feature group”; paragraphs [0051], “Once the predictive model is trained using the training data 208, the system 200 executes the trained predictive model using the input data 212 to generate test data 216”; test data is the prediction outputs of the predictive model; paragraphs [0063]-[0064], “The model correction 224 receives the identification of the cause of the bias and generates a model correction that, when applied to the predictive model, reduces or eliminates the bias. The model correction includes modified training data, instructions to modify training data, a modification to the predictive model, or instructions to modify the predictive model… In some instances, the model correction can include modifications to the predictive model… the model correction 224 generates a model correction that modifies a portion of the predictive model that is associated with the bias”; alternative model is analogous to the version of the model in the reference that is modified for the purposes of removing bias; paragraph [0065], “If the model correction includes a modification to the predictive model 204, the system 200 applies the model correction to the predictive model 204… In other instances, the process repeats to determine of the model correction successfully reduced or eliminated the identified and if the model correction introduced a new bias”; if model itself is modified, it may be trained again on the training data (training data is considered analogous to production data));
to present an alternative prediction that is more accurate than the prediction of the in-production machine learning model. (paragraph [0089], “FIG. 6 depicts an example of a graphical user interface presenting bias analysis data, according to certain embodiments of the present disclosure. A bias detection system generates a graphical user interface to visually represent the biases in a predictive model”; paragraph [0091], “A second subpanel 612 indicates the overall accuracy of the predictive model in generating predictions. A third subpanel 616 represents four performance metrics associated with a particular prediction associated with the feature group. Returning to the employee evaluation example, a feature group of Asian males has three possible three possible predictions: below expectations, meets expectations, exceeds expectations. The subpanel 616 displays four performance metrics associated with the meets expectation prediction. The four performance metrics indicate the classification accuracy 620, classification rate 624, true positive rate 628, and false positive rate 632 for Asian males predicted to meet expectations” the alternative prediction exceeds expectations).
But Lee fail to explicitly disclose “wherein the machine learning model is an in-production machine learning model; determine that the in-production machine learning model is not performing as required for a particular category of data that corresponds to the first bin based on an average difference between the first set of residual values and the second set of residual values satisfies a residual threshold, train, based on determining that the in-production machine learning model is not performing as required for the particular category of data, an alternative machine learning model, for the particular category of data, using an updated training dataset that includes a portion of the training dataset that corresponds to the particular category of data and that is smaller than an entirety of the training dataset, wherein the portion of the training dataset includes the training set of entries;”
Grossman disclose “wherein the machine learning model is an in-production machine learning model; (col. 5, line 5-20, col. 7, line 46-col, 8, line 20, ML model may be deployed into a production env.)
determine that the in-production machine learning model is not performing as required for a particular category of data that corresponds to the first bin based on an average difference between the first set of residual values and the second set of residual values satisfies a residual threshold, (col. 2, line 26-43, col. 17, line 28-col. 23, line 32, col. 19, line 28-54, determine the ML model is not satisfying the threshold for the different sets values for the bin based on an average drift difference, the performance is based on each input variable that corresponding to each bin, and each input variable is corresponding to a type of data.)
train, based on determining that the in-production machine learning model is not performing as required for the particular category of data, an alternative machine learning model, for the particular category of data, using an updated training dataset that includes a portion of the training dataset that corresponds to the particular category of data and that is smaller than an entirety of the training dataset, wherein the portion of the training dataset includes the training set of entries; (col. 2, line 26-43, col. 17, line 28-col. 23, line 32, col. 19, line 28-54, train, based on the ML model is not satisfying the threshold, the performance is based on each input variable that corresponding to each bin, and each input variable is corresponding to a type of data, updating the ML using a different training set which is part of the first bin, such as my specifying different bin ranges, different value ranges, the portion of the training set can include the training set of entries, etc. Note: (1) please further specify the particular category of data since it is broad, and the relationships between value ranges, category of data and bin, etc. (2) make all the independent claims corresponding to each other to help move forward the prosecution, call to discuss if necessary)
It would have been obvious to one having ordinary skill in the art before the effective filing data of the claimed invention was made to incorporate Grossman‘s generating an alert regarding a ML model into Lee’s invention as they are related to the same field endeavor of model training and learning. The motivation to combine these arts, as proposed above, at least because Grossman’s ML model performance evaluation would help to provide more ML model evaluation method into Lee’s system. Therefore it would have been obvious to one having ordinary skill in the art before the effective filing data of the claimed invention was made that providing model performance evaluation in training ML model would help to improve accuracy of prediction precision and training efficiency.
But Lee and Grossman fail to explicitly disclose “
In regard to claim 21, claim 21 is a method claim corresponding to the system claim 1 above and, therefore, is rejected for the same reasons set forth in the rejections of claim 1.
In regard to claim 22, claim 22 is a medium claim corresponding to the system claim 1 above and, therefore, is rejected for the same reasons set forth in the rejections of claim 1.
Regarding claim 2
Regarding claim 2, Lee teaches the system of claim 1 (as mentioned above), wherein the instruction for generating the plurality of bins for the compare dataset, when executed by the one or more processors, further cause the one or more processors to:
select a first feature within the compare dataset (paragraph [0038], “The bias detection module 124 enumerates the features over which the predictive model is tested. In some instances, the bias detection module 124 enumerates the features to test the predictive model from an analysis of the training data 116 and/or the test data 120. For instance, the bias detection module 124 enumerates a set of features by identifying types of predictions that are generated, characteristics of a class (e.g., employees) that are the subject of the generated predictions, and/or statistically relevant aspects of the training data 208 or input data 212 such as data types, data structures, reoccurring data values, and the like”; enumeration allows the bias detection module to identify particular features of interest);
generate a plurality of bin labels for the first feature, wherein each bin is associated with a subset of values for the first feature (paragraph [0031], “For example, for a predictive model generating predictions of employee performance, the employees are representative of objects and a feature group corresponds to a group of employees that each share a common set of characteristics such as demographic information (e.g., Asian males), positional information (e.g., executive officers), geographical information (e.g., in the New York office, in the Northeast, etc.), salary information, historical predictions, combinations thereof, or the like”); and
assign entries of the compare dataset to the plurality of bin labels according to a corresponding value for the first feature for each entry in the compare dataset (paragraph [0043], “the model test node 104 determines the portion of the training data 116 that corresponds to the feature group female executives. The model test node 104 analysis the training data 116 to determine a type model correction to reduce the bias such as increasing training data corresponding to female executives with labels that correspond to meets expectations and exceeds expectations, decreasing training data associated with female executives that have a label of below expectations, increasing or decreasing training data corresponding to another feature or feature groups, or the like”).
Regarding claim 5
Regarding claim 5, Lee teaches a method comprising:
subsequent to a machine learning model being trained using a first dataset, generating a plurality of groups for a second dataset based on a parameter within the second dataset, (paragraph [0066]-[0067], “A block 304, a computing device receives data associated with a trained predictive model. The data includes data output from the predictive model such as predictions associated with a class or a subclass. In some instances, the data includes the input data that was input to the predictive model to generate the output and training data used during a training phase of the predictive model… At block 308, the computing device identifies a feature group from the data”; paragraph [0054], “A feature group is made up of one or more features and represents a subclass of the class that includes members that each include characteristics that correspond to the one or more features”; feature groups can be considered analogous to parameters, and they may contain one or more features that intuitively correspond to groups of values);
obtaining a first set of entries corresponding to a group from the first dataset and a second set of entries corresponding to the group from the second dataset;(paragraph [0042], “In some instances, model test node 104 receives a seed from the training datasets 132… The model test node 104 uses the seed to procedurally generate any amount of training data that corresponds to the predictive model. The seed may be combined with a particular feature or feature group to generate training data that corresponds to that feature or feature group”; paragraph [0038], “For instance, the bias detection module 124 enumerates a set of features by identifying types of predictions that are generated, characteristics of a class (e.g., employees) that are the subject of the generated predictions, and/or statistically relevant aspects of the training data 208 or input data 212 such as data types, data structures, reoccurring data values, and the like”; input data is analogous to the second dataset in this example; model test node generates (“obtains”) training data and bias detection module takes input data from feature sets);
inputting the first set of entries corresponding to the group into the machine learning model to obtain a first set of error values; (paragraph [0060], “The results of bias detection module 220 includes, but is not limited to, an identification of a bias associated with a feature group, a performance characteristic of the predictive model, an error, combinations thereof, an identification of a cause of the bias or error such as a characteristic of the predictive model 204, a portion of the training data 208, a portion of the input data 212, combinations thereof, or the like”; training data given to predictive model and error is output);
inputting the second set of entries into the machine learning model to obtain a second set of error values; (paragraph [0060], “The results of bias detection module 220 includes, but is not limited to, an identification of a bias associated with a feature group, a performance characteristic of the predictive model, an error, combinations thereof, an identification of a cause of the bias or error such as a characteristic of the predictive model 204, a portion of the training data 208, a portion of the input data 212, combinations thereof, or the like”; input data given to predictive model and error is output); and
But Lee fail to explicitly disclose “wherein the machine learning model is an in-production machine learning model; wherein the machine learning model is an in-production machine learning model; determine that the in-production machine learning model is not performing as required for particular category of data that corresponds to the group based on a difference between the first set of error values and the second set of error values satisfying a threshold; in response to determining the in-production machine learning model is not performing as required for the particular category of data, transmitting a notification.”
Grossman disclose “wherein the machine learning model is an in-production machine learning model; wherein the machine learning model is an in-production machine learning model; (col. 5, line 5-20, col. 7, line 46-col, 8, line 20, ML model may be deployed into a production env.)
determine that the in-production machine learning model is not performing as required for particular category of data that corresponds to the group based on a difference between the first set of error values and the second set of error values satisfying a threshold; in response to determining the in-production machine learning model is not performing as required for the particular category of data, transmitting a notification. (col. 2, line 26-43, col. 17, line 28-col. 23, line 32, col. 19, line 28-54, determine the ML model is not satisfying the threshold, the performance is based on each input variable that corresponding to each bin, and each input variable is corresponding to a type of data, based on an average drift difference and sending an alert to the user indicating the dataset)
It would have been obvious to one having ordinary skill in the art before the effective filing data of the claimed invention was made to incorporate Grossman‘s generating an alert regarding a ML model into Lee’s invention as they are related to the same field endeavor of model training and learning. The motivation to combine these arts, as proposed above, at least because Grossman’s ML model performance evaluation would help to provide more ML model evaluation method into Lee’s system. Therefore it would have been obvious to one having ordinary skill in the art before the effective filing data of the claimed invention was made that providing model performance evaluation in training ML model would help to improve accuracy of prediction precision and training efficiency.
Regarding claim 6
Regarding claim 6, Lee teaches the method of claim 5 (as explained above), wherein generating the plurality of groups comprises:
selecting a first parameter within the first dataset, wherein the first parameter corresponds to a set of parameter values (paragraph [0038], “The bias detection module 124 enumerates the features over which the predictive model is tested. In some instances, the bias detection module 124 enumerates the features to test the predictive model from an analysis of the training data 116 and/or the test data 120. For instance, the bias detection module 124 enumerates a set of features by identifying types of predictions that are generated, characteristics of a class (e.g., employees) that are the subject of the generated predictions, and/or statistically relevant aspects of the training data 208 or input data 212 such as data types, data structures, reoccurring data values, and the like”; enumeration allows the bias detection module to identify particular features of interest and group them into feature groups); and
generating a plurality of group definitions for the set of parameter values. (paragraph [0031], “For example, for a predictive model generating predictions of employee performance, the employees are representative of objects and a feature group corresponds to a group of employees that each share a common set of characteristics such as demographic information (e.g., Asian males), positional information (e.g., executive officers), geographical information (e.g., in the New York office, in the Northeast, etc.), salary information, historical predictions, combinations thereof, or the like”).
Regarding claim 7
Regarding claim 7, Lee teaches the method of claim 6 (as explained above), wherein obtaining the first set of entries from the first dataset comprises:
obtaining the first set of entries from the first dataset comprises: determining, for each entry in the first dataset, a group definitions of the plurality of group definition that matches a parameter value of the set of parameter values; and assigning each entry of the first dataset to a group of the plurality of groups based on determining the group definition. (paragraph [0041], “The bias detection module 124 identifies one or more types of bias associated with a particular feature or feature group then identifies the portion of the training data 116 that corresponds to the feature or feature group”; feature groups are well-defined according to paragraph [0031]; identifying training data corresponding to feature groups is indicative of matching data entries to group definitions; training data entries are considered assigned to the feature group they belong to).
Regarding claim 13
Regarding claim 13, Lee teaches:
A non-transitory, computer-readable media comprising instructions that when executed by one or more processors, cause the one or more processors to perform operations comprising (paragraph [0129], “The computing system 1000 executes program code that configures the processor 1102 to perform one or more of the operations described herein. The program code may be resident in the memory device 1008 or any suitable computer-readable medium and may be executed by the processor 1004 or any other suitable processor”; “memory device 1008” implies the use of tangible hardware, therefore a non-transitory computer-readable storage medium);
subsequent to a machine learning model being trained using a first dataset, generating a plurality of groups based on a first dataset, (paragraph [0066]-[0067], “A block 304, a computing device receives data associated with a trained predictive model. The data includes data output from the predictive model such as predictions associated with a class or a subclass. In some instances, the data includes the input data that was input to the predictive model to generate the output and training data used during a training phase of the predictive model… At block 308, the computing device identifies a feature group from the data”; paragraph [0054], “A feature group is made up of one or more features and represents a subclass of the class that includes members that each include characteristics that correspond to the one or more features”; feature groups can be considered analogous to parameters, and they may contain one or more features that intuitively correspond to groups of values);
But Lee fail to explicitly disclose “wherein the machine learning model is an in-production machine learning model; determine that the in-production machine learning model is not performing as required for a particular category of data that corresponds to a group of the plurality of groups; in response to determining the in-production machine learning model is not performing as required for the particular category of data, transmitting a notification.”
Grossman disclose “wherein the machine learning model is an in-production machine learning model; (col. 5, line 5-20, col. 7, line 46-col, 8, line 20, ML model may be deployed into a production env.)
determine that the in-production machine learning model is not performing as required for a particular category of data that corresponds to a group of the plurality of groups; in response to determining the in-production machine learning model is not performing as required for the particular category of data, transmitting a notification. (col. 2, line 26-43, col. 17, line 28-col. 23, line 32, col. 19, line 28-54, determine the ML model is not satisfying the threshold, the performance is based on each input variable that corresponding to each bin, and each input variable is corresponding to a type of data, based on an average drift difference and sending an alert to the user indicating the dataset)
It would have been obvious to one having ordinary skill in the art before the effective filing data of the claimed invention was made to incorporate Grossman‘s generating an alert regarding a ML model into Lee’s invention as they are related to the same field endeavor of model training and learning. The motivation to combine these arts, as proposed above, at least because Grossman’s ML model performance evaluation would help to provide more ML model evaluation method into Lee’s system. Therefore it would have been obvious to one having ordinary skill in the art before the effective filing data of the claimed invention was made that providing model performance evaluation in training ML model would help to improve accuracy of prediction precision and training efficiency.
Regarding claim 14
Regarding claim 14, Lee teaches the one or more non-transitory, computer-readable media of claim 13 (as explained above), wherein the instructions for generating the plurality of groups further cause the one or more processors to perform operations comprising:
selecting a first parameter within the first dataset, wherein the first parameter corresponds to a set of parameter values (paragraph [0038], “The bias detection module 124 enumerates the features over which the predictive model is tested. In some instances, the bias detection module 124 enumerates the features to test the predictive model from an analysis of the training data 116 and/or the test data 120. For instance, the bias detection module 124 enumerates a set of features by identifying types of predictions that are generated, characteristics of a class (e.g., employees) that are the subject of the generated predictions, and/or statistically relevant aspects of the training data 208 or input data 212 such as data types, data structures, reoccurring data values, and the like”; enumeration allows the bias detection module to identify particular features of interest and group them into feature groups); and
generate a plurality of group definitions for set of parameters values, wherein each group definition is associated with a subset of the set of parameter values (paragraph [0031], “For example, for a predictive model generating predictions of employee performance, the employees are representative of objects and a feature group corresponds to a group of employees that each share a common set of characteristics such as demographic information (e.g., Asian males), positional information (e.g., executive officers), geographical information (e.g., in the New York office, in the Northeast, etc.), salary information, historical predictions, combinations thereof, or the like”).
Regarding claim 15
Regarding claim 15, Lee teaches the one or more non-transitory, computer-readable media of claim 14 (as explained above), wherein the instructions further cause the one or more processors to perform operations comprising:
determining, for each entry in the first dataset, a group definition of the plurality of group definitions that matches a parameter value; and assigning each entry of the first dataset to a group of the plurality of groups based on determining the group definition. (paragraph [0041], “The bias detection module 124 identifies one or more types of bias associated with a particular feature or feature group then identifies the portion of the training data 116 that corresponds to the feature or feature group”; feature groups are well-defined according to paragraph [0031]; identifying training data corresponding to feature groups is indicative of matching data entries to group definitions; training data entries are considered assigned to the feature group they belong to).
Regarding claim 23
Regarding claim 23, dependent on claim 5, the rejection of claim 1 is incorporated. In addition, But Lee fail to explicitly disclose “wherein the particular category of data is a range of parameter of values for a particular parameter.”
Grossman disclose “wherein the particular category of data is a range of parameter of values for a particular parameter.” (col. 2, line 26-43, col. 17, line 28-col. 23, line 32, col. 19, line 28-54, (and each input variable is corresponding to a type of data and have range values)
It would have been obvious to one having ordinary skill in the art before the effective filing data of the claimed invention was made to incorporate Grossman‘s generating an alert regarding a ML model into Lee’s invention as they are related to the same field endeavor of model training and learning. The motivation to combine these arts, as proposed above, at least because Grossman’s ML model performance evaluation would help to provide more ML model evaluation method into Lee’s system. Therefore it would have been obvious to one having ordinary skill in the art before the effective filing data of the claimed invention was made that providing model performance evaluation in training ML model would help to improve accuracy of prediction precision and training efficiency.
Claim 4 is rejected under 35 U.S.C. 103 as being unpatentable over Lee and Grossman US 12165017 in view of in view of Lopes et al. (“Machine Learning System for Transaction Reconciliation”—US 20210350382 A1; hereinafter “Lopes”) and Fukushima et al. (“Information Processing Apparatus, Information Processing Method, And Computer Program”—U.S. Patent No. 12,136,023; hereinafter “Fukushima”).
Regarding claim 4, dependent on claim 1, the rejection of claim 1 is incorporated. In addition, Lee teaches The system of claim 1 wherein the instructions when executed by the one or more processors, further cause the one or more processors to… determining a base residual score for the first set of entries and a compare residual score for the second set of entries (Lee, paragraph [0060], “The results of bias detection module 220 are passed to model corrections 224. The results of bias detection module 220 includes, but is not limited to, an identification of a bias associated with a feature group, a performance characteristic of the predictive model, an error, combinations thereof, an identification of a cause of the bias or error such as a characteristic of the predictive model 204, a portion of the training data 208, a portion of the input data 212, combinations thereof, or the like”).
Lee and Grossman do not teach the limitation wherein the instructions when executed by the one or more processors, further cause the one or more processors to: calculating the average difference between the first set of residual values and the second set of residual values based on the base residual score and the compare residual score. However, Lopes teaches wherein the instructions when executed by the one or more processors, further cause the one or more processors to: calculating the residual values based on the base residual score and the compare residual score (Lopes, paragraph [0059], “For example, the transaction reconciliation system may determine a first error metric associated with the set of matched entries, may determine a second error metric associated with the set of classified entries, may compare the two error metrics, and may perform an action based on the comparison. For example, the transaction reconciliation system may perform an action based on a difference between the error metrics, whether the difference satisfies a condition, whether the difference satisfies a threshold, and/or the like”), while Fukushima teaches calculating the average of first and second sets of residual values (Fukushima, paragraph (57) and (58), “Examples of calculating, as the improved score, an aggregation of errors that is a divided group include the following examples 3 and 4… where the average of the errors of the first group is… and the average of the errors of the second group is…”; residual values are considered error values; and while Fukushima teaches averaging two sets of errors between which a difference can be found as taught by Lopes, it can be shown mathematically that the difference of two averages is equivalent to averaging a corresponding set of differences).
Lopes and Fukushima are considered analogous to the claimed invention since they both utilize error values in machine learning. It would have been obvious to a PHOSITA, before the effective filing date of the claimed invention, to incorporate the methodologies of Lopes and Fukushima into those of Grossman and Lee. Taking an average of error values (as is frequently done with statistical measures such as mean absolute error) is considered combining prior art elements in known ways to yield predictable results according to MPEP 2141(III).
Claims 8-11, and 16-19 are rejected under 35 U.S.C. 103 as being unpatentable over Lee et al. (“Methods and Systems for Detection and Isolation of Bias in Predictive Models”—U.S. Patent No. 11,593,648; hereinafter “Lee”) and Grossman US 12165017 in view of Rama et al. (“Apparatus and Method For Anomaly Detection Using Weighted Autoencoder”—US 20220198267 A1; hereinafter “Rama”).
Claims 8, and 16
Regarding claim 8
Regarding claim 8, dependent on claim 5, the rejection of claim 5 is incorporated. In addition, Lee teaches The method of claim 5 further comprising: determining a type of data associated with the group (Lee, paragraph [0038], “For instance, the bias detection module 124 enumerates a set of features by identifying types of predictions that are generated, characteristics of a class (e.g., employees) that are the subject of the generated predictions, and/or statistically relevant aspects of the training data 208 or input data 212 such as data types, data structures, reoccurring data values, and the like”).
Lee and Grossman do not teach the limitations further comprising… sorting, according to the type of data, parameter values associated with the group, and assigning entries of the first dataset to the plurality of groups according to the sorting. However, Rama teaches the limitations further comprising… sorting, according to the type of data and parameter values associated with the group (Rama, paragraph [0029], “The feature values of each observation are grouped into bins such that there are not overlapping intervals between the bins. The feature values of the observations are first sorted in ascending order and divided into buckets, i.e., bins, such that each bin has an equal number of observations. This is shown as Stage 1 in which each bin has three observations with the smallest feature value at the top. Let featureistart and featureiend denote the starting and ending values of each of the i groups. All overlapping intervals with the same value are merged into the same interval as shown in Stage 2. This is also represented by Equation (2) below, where b denotes the number of bins”, as well as assigning entries of the first dataset to the plurality of groups according to the sorting (Rama, paragraph [0029], “FIG. 3 is a diagram of one example of an interval width binning heuristic that may be applied to the input training observations in order to determine weights for the autoencoder…”). It would have been obvious to PHOSITA, before the effective filing date of the claimed invention, to incorporate the methodologies of Rama into those of Grossman and Lee. Sorting a list of items prior to their processing is a technique commonly employed by developers of software and AI models, and would be considered combining prior art elements according to known methods to yield predictable results as per MPEP 2141(III).
Regarding claim 16
Regarding claim 16, dependent on claim 13, the rejection of claim 13 is incorporated. In addition, Lee teaches The one or more non-transitory computer-readable media of claim 13, wherein the instructions further cause the one or more processors to perform operations comprising: determining a type of data associated with the group (Lee, paragraph [0038], “For instance, the bias detection module 124 enumerates a set of features by identifying types of predictions that are generated, characteristics of a class (e.g., employees) that are the subject of the generated predictions, and/or statistically relevant aspects of the training data 208 or input data 212 such as data types, data structures, reoccurring data values, and the like”).
Lee and Grossman do not teach the limitations the instructions further causing the one or more processors to perform operations comprising… sorting, according to the type of data, parameter values associated with the group, and assigning entries of the first dataset to the plurality of groups according to the sorting. However, Rama teaches the limitations the instructions further causing the one or more processors to perform operations comprising: sorting, according to the type of data and parameter value associated with the first group (Rama, paragraph [0029], “The feature values of each observation are grouped into bins such that there are not overlapping intervals between the bins. The feature values of the observations are first sorted in ascending order and divided into buckets, i.e., bins, such that each bin has an equal number of observations. This is shown as Stage 1 in which each bin has three observations with the smallest feature value at the top. Let featureistart and featureiend denote the starting and ending values of each of the i groups. All overlapping intervals with the same value are merged into the same interval as shown in Stage 2. This is also represented by Equation (2) below, where b denotes the number of bins” as well as assigning entries of the first dataset to the plurality of groups according to the sorting (Rama, paragraph [0029], “FIG. 3 is a diagram of one example of an interval width binning heuristic that may be applied to the input training observations in order to determine weights for the autoencoder…”). It would have been obvious to PHOSITA, before the effective filing date of the claimed invention, to incorporate the methodologies of Rama into those of Grossman and Lee. Sorting a list of items prior to their processing is a technique commonly employed by developers of software and AI models, and would be considered combining prior art elements according to known methods to yield predictable results as per MPEP 2141(III).
Claims 9, 11, 17, and 19
Claims 9, 11, 17, and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Lee, Grossman in view of Lopes.
Regarding claim 9
Regarding claim 9, dependent on claim 5, the rejection of claim 5 is incorporated. In addition, Lee teaches The method of claim 5 further comprising: determining a first error score for the first set of entries and a second error score for the second set of entries (paragraph [0060], “The results of bias detection module 220 are passed to model corrections 224. The results of bias detection module 220 includes, but is not limited to, an identification of a bias associated with a feature group, a performance characteristic of the predictive model, an error, combinations thereof, an identification of a cause of the bias or error such as a characteristic of the predictive model 204, a portion of the training data 208, a portion of the input data 212, combinations thereof, or the like”).
Lee does not teach the limitation further comprising… calculating the difference based on the first error score and the second error score. However, Lopes teaches this limitation (paragraph [0059], “For example, the transaction reconciliation system may determine a first error metric associated with the set of matched entries, may determine a second error metric associated with the set of classified entries, may compare the two error metrics, and may perform an action based on the comparison. For example, the transaction reconciliation system may perform an action based on a difference between the error metrics, whether the difference satisfies a condition, whether the difference satisfies a threshold, and/or the like. As a specific example, if the second error metric indicates a greater error than the first error metric and/or if the difference between the error metrics satisfies a threshold, then the transaction reconciliation system may predict missing transaction data”). It would have been obvious to a PHOSITA, before the effective filing date of the claimed invention, to incorporate the methodologies of Lopes into those of Grossman and Lee. Calculating a difference between sets of error values is considered combining prior art elements in known ways to yield predictable results according to MPEP 2141(III).
Regarding claim 11
Regarding claim 11, dependent on claim 9, the rejection of claim 9 is incorporated. In addition, the combination of Lee, Grossman and Lopes teaches the limitation The method of claim 9 further comprising determining the difference between the first set error of values and the second set of error values based on the second error score being higher than the first error score by an error threshold. In particular, Lopes teaches this limitation (Lopes, paragraph [0059], “For example, the transaction reconciliation system may determine a first error metric associated with the set of matched entries, may determine a second error metric associated with the set of classified entries, may compare the two error metrics, and may perform an action based on the comparison. For example, the transaction reconciliation system may perform an action based on a difference between the error metrics, whether the difference satisfies a condition, whether the difference satisfies a threshold, and/or the like. As a specific example, if the second error metric indicates a greater error than the first error metric and/or if the difference between the error metrics satisfies a threshold, then the transaction reconciliation system may predict missing transaction data”). It would have been obvious to a PHOSITA, before the effective filing date of the claimed invention, to incorporate the methodologies of Lopes into those of Grossman and Lee. Calculating a difference between sets of error values is considered combining prior art elements in known ways to yield predictable results according to MPEP 2141(III).
Regarding claim 17
Regarding claim 17, dependent on claim 13, the rejection of claim 13 is incorporated. In addition, Lee teaches The one or more non-transitory, computer-readable media of claim 13, wherein the instructions further cause the one or more processors to perform operations comprising: determining a first error score for a first set of entries, corresponding to the group and a second error score for a second set of entries corresponding to the group; (paragraph [0060], “The results of bias detection module 220 are passed to model corrections 224. The results of bias detection module 220 includes, but is not limited to, an identification of a bias associated with a feature group, a performance characteristic of the predictive model, an error, combinations thereof, an identification of a cause of the bias or error such as a characteristic of the predictive model 204, a portion of the training data 208, a portion of the input data 212, combinations thereof, or the like”).
Lee and Grossman do not teach the limitation the instructions further causing the one or more processors to perform operations comprising… calculating a difference based on the first error score and the second error score. However, Lopes teaches this limitation (paragraph [0059], “For example, the transaction reconciliation system may determine a first error metric associated with the set of matched entries, may determine a second error metric associated with the set of classified entries, may compare the two error metrics, and may perform an action based on the comparison. For example, the transaction reconciliation system may perform an action based on a difference between the error metrics, whether the difference satisfies a condition, whether the difference satisfies a threshold, and/or the like. As a specific example, if the second error metric indicates a greater error than the first error metric and/or if the difference between the error metrics satisfies a threshold, then the transaction reconciliation system may predict missing transaction data”). It would have been obvious to a PHOSITA, before the effective filing date of the claimed invention, to incorporate the methodologies of Lopes into those of Grossman and Lee. Calculating a difference between sets of error values is considered combining prior art elements in known ways to yield predictable results according to MPEP 2141(III).
Regarding claim 19
Regarding claim 19, dependent on claim 17, the rejection of claim 17 is incorporated. In addition, the combination of Lee, Grossman and Lopes teaches the limitation The one or more non-transitory, computer-readable media of claim 17, wherein the instructions further cause the one or more processors perform operations comprising determining a difference based on the second error score being higher than the first error score. In particular, Lopes teaches this limitation (Lopes, paragraph [0059], “For example, the transaction reconciliation system may determine a first error metric associated with the set of matched entries, may determine a second error metric associated with the set of classified entries, may compare the two error metrics, and may perform an action based on the comparison. For example, the transaction reconciliation system may perform an action based on a difference between the error metrics, whether the difference satisfies a condition, whether the difference satisfies a threshold, and/or the like. As a specific example, if the second error metric indicates a greater error than the first error metric and/or if the difference between the error metrics satisfies a threshold, then the transaction reconciliation system may predict missing transaction data”). It would have been obvious to a PHOSITA, before the effective filing date of the claimed invention, to incorporate the methodologies of Lopes into those of Grossman and Lee. Calculating a difference between sets of error values is considered combining prior art elements in known ways to yield predictable results according to MPEP 2141(III).
Claims 10 and 18
Claims 10 and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Lee and Grossman in view of Lopes as applied to claims 5 and 13 above, and further in view of Fukushima.
Regarding claim 10
Regarding claim 10, dependent on claim 9, the rejection of claim 9 is incorporated. In addition, the combination of Lee, Grossman and Lopes teaches the limitation The method of claim 9 but does not teach wherein determining the first error score for the first set of entries and the second error score for the second set of entries comprises generating a first average value for the first set of error values and a second average value for the second set of error values. However, Fukushima teaches this limitation (Fukushima, paragraph (57) and (58), “Examples of calculating, as the improved score, an aggregation of errors that is a divided group include the following examples 3 and 4… where the average of the errors of the first group is… and the average of the errors of the second group is…”). It would have been obvious to a PHOSITA, before the effective filing date of the claimed invention, to incorporate the teachings of Fukushima into those of Grossman, Lee and Lopes. Taking the average of multiple error values is common practice in statistical settings to determine, for example, a mean squared error. This is considered combining prior art elements in known ways to yield predictable results according to MPEP 2141(III).
Regarding claim 18
Regarding claim 18, dependent on claim 17, the rejection of claim 17 is incorporated. In addition, the combination of Lee, Grossman and Lopes teaches the limitation The one or more non-transitory computer-readable media of claim 17 but does not teach wherein the instructions for determining the first error score for the first set of entries and the second error score for the second set of entries comprises generating a first average value for a first set of error values obtained from the machine learning model and a second average value for a second set of error values from the machine learning model. However, Fukushima teaches this limitation (Fukushima, paragraph (57) and (58), “Examples of calculating, as the improved score, an aggregation of errors that is a divided group include the following examples 3 and 4… where the average of the errors of the first group is… and the average of the errors of the second group is…”). It would have been obvious to a PHOSITA, before the effective filing date of the claimed invention, to incorporate the teachings of Fukushima into those of Grossman, Lee and Lopes. Taking the average of multiple error values is common practice in statistical settings to determine, for example, a mean squared error. This is considered combining prior art elements in known ways to yield predictable results according to MPEP 2141(III).
Response to Arguments
Applicant’s arguments filed on 7/30/2026 have been considered but are moot because the arguments do not apply to the current rejection.
Conclusion
The prior art made of record and not relied upon is considered pertinent to Applicant's disclosure.
PATENT PUB. # PUB. DATE INVENTOR(S) TITLE
US20210065053 2021-03-04 HIGGINS et al.
AUTOMATED DATA PROCESSING AND MACHINE LEARNING MODEL GENERATION
HIGGINS et al. disclose A device may obtain first data relating to a machine learning model. The device may pre-process the first data to alter the first data to generate second data. The device may process the second data to select a set of features from the second data. The device may analyze the set of features to evaluate a plurality of types of machine learning models with respect to the set of features. The device may select a particular type of machine learning model for the set of features based on analyzing the set of features to evaluate the plurality of types of machine learning models. The device may tune a set of parameters of the particular type of machine learning model to train the machine learning model. The device may receive third data for prediction. The device may provide a prediction using the particular type of machine learning model… see abstract.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to XUYANG XIA whose telephone number is (571)270-3045. The examiner can normally be reached Monday-Friday 8am-4pm.
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, Jennifer Welch can be reached at 571-272-7212. 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.
XUYANG XIA
Primary Examiner
Art Unit 2143
/XUYANG XIA/Primary Examiner, Art Unit 2143
Prosecution Insights