Feature selection using feedback-assisted optimization models

Patent metadata

Field	Value
Publication number	US-12579477-B2
Application number	US-202117475901-A
Country	US
Kind code	B2
Filing date	Sep 15, 2021
Priority date	May 11, 2021
Publication date	Mar 17, 2026
Grant date	Mar 17, 2026

Abstract

Techniques are disclosed relating to feature selection based on feedback-assisted optimization models. In various embodiments, for example, the disclosed techniques include accessing a training dataset that includes a plurality of data samples that include data values for a plurality of features, and a set of labels corresponding to the plurality of data samples. In some embodiments, a computer system performs feature-selection operations to select, from the plurality of features, a subset of features to include in a reduced feature set. For example, in some embodiments the feature-selection operations include processing the training dataset based on an optimization model, where an objective function utilized in the optimization model utilizes performance feedback information corresponding to machine learning models that are trained based on candidate feature sets. Based on the feature-selection operation, the computer system may generate an output value that indicates the subset of features to include in the reduced feature set.

First claim

What is claimed is:

1 . A method, comprising:
accessing, by a computer system, a training dataset that includes:
a plurality of data samples that include data values for a plurality of features; and
a set of labels corresponding to the plurality of data samples;
performing, by the computer system, a feature-selection operation to select, from the plurality of features, a subset of features to include in a reduced feature set, wherein the feature-selection operation includes:
executing trained machine learning models that are trained based on candidate feature subsets;
generating performance feedback information based on output of the trained machine learning models for the candidate feature subsets, wherein the performance feedback information indicates at least variance in an accuracy of the trained machine learning models across the candidate feature subsets; and
processing the training dataset using quantum computing based on an optimization model, including executing an objective function, utilized in the optimization model, using at least the performance feedback information; and
generating, by the computer system, an updated training dataset that includes only the subset of features, wherein the generating is performed based on the feature-selection operation:
applying a penalty value to the objective function in response to a determination that the variance in accuracy exceeds a particular threshold value; and
indicating ground state spin information determined using the quantum computing; and

evaluating a risk associated with a requested operation by executing the trained machine learning models to obtain an output; and denying the requested operation or requesting user authentication operations based on the output.

2 . The method of claim 1 , wherein the optimization model is a quadratic unconstrained binary optimization (“QUBO”) model, wherein the output includes ground state spin information that corresponds to a minimization of the objective function utilized in the QUBO model.

3 . The method of claim 1 , wherein performing the feature-selection operation further includes:
identifying a first candidate feature set based on a first iteration of the feature-selection operation; receiving first performance feedback information corresponding to a performance of a first machine learning model trained based on the first candidate feature set; and based on the first performance feedback information, modifying a penalty term in the objective function.

4 . The method of claim 1 , wherein performing the feature-selection operation further includes:
calculating log-loss scores for the trained machine learning models based on output of the trained machine learning models across a plurality of test splits; and calculating, based on the log-loss scores, the variance in the accuracy of the one or more of the trained machine learning models.

5 . The method of claim 3 ,
wherein the method further comprises performing an additional iteration of the feature-selection operation to select the reduced feature set.

6 . The method of claim 1 , wherein the processing the training dataset based on the optimization model includes selecting the subset of features that maximizes a measure of relevancy between pairs of the plurality of features and the set of labels for the plurality of data samples.

7 . The method of claim 1 , wherein processing the training dataset based on the optimization model includes using multi-objective optimization based on goal programming, including by:
dividing the objective function into a plurality of goals; optimizing a first one of the plurality of goals; and based on the optimization of the first one of the plurality of goals, optimizing one or more additional goals from the plurality of goals in the objective function.

8 . The method of claim 1 , wherein processing the training dataset based on the optimization model includes using multi-objective optimization based on Pareto optimal sets, including by:
analyzing a plurality of goals of the objective function; and identifying a plurality of Pareto fronts having corresponding performances across the plurality of goals.

9 . The method of claim 8 , wherein the plurality of goals includes one or more of the following goals:
minimization of redundancy between features in the reduced feature set; maximization of relevance between the features in the reduced feature set; and reduction of variance in test scores generated during a testing of the reduced feature set using a plurality of different test splits.

10 . The method of claim 1 , further comprising:
generating, by the computer system, an updated training dataset that includes data values for the subset of features that are included in the reduced feature set, wherein the updated training dataset does not include second data values for one or more of the plurality of features that are not included in the reduced feature set; and training, by the computer system, a machine learning model based on the updated training dataset.

11 . A non-transitory, computer-readable medium having instructions stored thereon that are executable by a computer system to perform operations comprising:
accessing a training dataset that includes: a plurality of data samples that include data values for a plurality of features; and a set of labels corresponding to the plurality of data samples; performing a feature-selection operation to select, from the plurality of features, a subset of features to include in a reduced feature set, wherein the feature-selection operation includes: executing trained machine learning models that are trained based on candidate feature subsets; generating performance feedback information based on output of the trained machine learning models for the candidate feature subsets, wherein the performance feedback information indicates at least variance in an accuracy of the trained machine learning models across the candidate feature subsets; and processing the training dataset using quantum computing based on an optimization model, including executing an objective function, utilized in the optimization model, using at least the performance feedback information; and generating an updated training dataset that includes only the subset of features, wherein the generating is performed based on the feature-selection operation:
applying a penalty value to the objective function in response to a determination that the variance in accuracy exceeds a particular threshold value; and
indicating ground state spin information determined using the quantum computing; and
evaluating a risk associated with a requested operation by executing the trained machine learning models to obtain an output; and denying the requested operation or requesting user authentication operations based on the output.

12 . The non-transitory, computer-readable medium of claim 11 , wherein performing the feature-selection operation further includes:
identifying a first candidate feature set based on a first iteration of the feature-selection operation; receiving first performance feedback information corresponding to a performance of a first machine learning model trained based on the first candidate feature set; and based on the first performance feedback information, modifying a penalty term in the objective function.

13 . The non-transitory, computer-readable medium of claim 12 , wherein the variance is determined based on log-loss scores for the first machine learning model across a plurality of test splits.

14 . The non-transitory, computer-readable medium of

Assignees

• Paypal Inc

Inventors

• Sharma Nitin S
• Kumar Niraj

Classifications

• G06F18/2113

  by ranking or filtering the set of features, e.g. using a measure of variance or of feature cross-correlation · CPC title

• G06F18/217

  Validation; Performance evaluation; Active pattern learning techniques · CPC title

• G06F18/214

  Generating training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06N10/00

  Quantum computing, i.e. information processing based on quantum-mechanical phenomena · CPC title

• G06N20/00

  Machine learning · CPC title