Patent Issued for Method of controlling for undesired factors in machine learning models (USPTO 11501133): State Farm Mutual Automobile Insurance Company

2022 DEC 01 (NewsRx) -- By a News Reporter-Staff News Editor at Insurance Daily News -- State Farm Mutual Automobile Insurance Company (Bloomington, Illinois, United States) has been issued patent number 11501133, according to news reporting originating out of Alexandria, Virginia, by NewsRx editors.

The patent’s inventors are Bernico, Michael L. (Bloomington, IL, US), Myers, Jeffrey S. (Normal, IL, US), Sanchez, Kenneth J. (San Francisco, CA, US).

This patent was filed on June 4, 2020 and was published online on November 15, 2022.

From the background information supplied by the inventors, news correspondents obtained the following quote: “Machine learning models may be trained to analyze information for particular purposes involving identifying correlations and making predictions. During training, the models may learn to include illegitimate, non-useful, irrelevant, misleading, or otherwise undesired factors, especially if such biases are present in the training data sets. In particular, while training with structured data involves limiting the data that a model considers, training with unstructured data allows the model to consider all available data, including background information and other undesired factors. For example, a neural network trained with unstructured data including people’s appearances to make correlations and predictions about those people may consider such undesired factors as age, sex, ethnicity, and/or race in its subsequent analyses.”

Supplementing the background information on this patent, NewsRx reporters also obtained the inventors’ summary information for this patent: “Embodiments of the present technology relate to machine learning models that control for consideration of one or more undesired factors which might otherwise be considered by the machine learning model when analyzing new data. For example, one embodiment of the present invention may be configured for training and using a neural network that controls for consideration of one or more undesired factors which might otherwise be considered by the neural network when analyzing new data as part of an underwriting process to determine an appropriate insurance premium.

“In a first aspect, a method of training and using a machine learning model that controls for consideration of one or more undesired factors which might otherwise be considered by the machine learning model may broadly comprise the following. The machine learning model may be trained using a training data set that contains information including the undesired factors. The undesired factors and one or more relevant interaction terms between the undesired factors may be identified. The machine learning model may then be caused to not consider the identified undesired factors when analyzing the new data to control for undesired prejudice or discrimination in machine learning models.

“In a second aspect, a computer-implemented method for training and using a machine learning model to evaluate an insurance applicant as part of an underwriting process to determine an appropriate insurance premium, wherein the machine learning model controls for consideration of one or more undesired factors which might otherwise be considered by the machine learning model, may broadly comprise the following. The machine learning model may be trained to probabilistically correlate an aspect of appearance with a personal and/or health-related characteristic by providing machine learning model with a training data set of images of individuals having known personal or health-related characteristics, including the undesired factors. The undesired factors and one or more relevant interaction terms between the undesired factors may be identified. An image of the insurance applicant may be received via a communication element. The machine learning model may analyze the image of the insurance applicant to probabilistically determine the personal and/or health-related characteristics for the insurance applicant, wherein such analysis excludes the identified undesired factors. The machine learning model may then suggest the appropriate insurance premium based at least in part on the probabilistically determined personal and/or health-related characteristic but not on the undesired factors.

“Various implementations of these aspects may include any one or more of the following additional features. Identifying the undesired factors and relevant interaction terms may include training a second machine learning model using a second training data set that contains only the undesired factors and the relevant interaction terms. Further, causing the machine learning model to not consider the identified undesired factors when analyzing the new data may include combining the machine learning model and the second machine learning model to eliminate a bias created by the undesired factors from the machine learning model’s consideration prior to employing the machine learning model to analyze the new data. Alternatively or additionally, identifying the undesired factors and relevant interaction terms may include training the machine learning model to identify the undesired factors and the one or more relevant interaction terms. Further, causing the machine learning model to not consider the identified undesired factors when analyzing the new data may include instructing the machine learning model to not consider the identified undesired factors while analyzing the new data. The machine learning model may be a neural network. The second machine learning model may be a linear model. The machine learning model may be trained to analyze the new data as part of an underwriting process to determine an appropriate insurance premium, and the new data may include images of a person applying for life insurance or health insurance or images of a piece of property for which a person is applying for property insurance. The machine learning model may be further trained to analyze the new data as part of the underwriting process to determine one or more appropriate terms of coverage.

“Advantages of these and other embodiments will become more apparent to those skilled in the art from the following description of the exemplary embodiments which have been shown and described by way of illustration. As will be realized, the present embodiments described herein may be capable of other and different embodiments, and their details are capable of modification in various respects. Accordingly, the drawings and description are to be regarded as illustrative in nature and not as restrictive.”

The claims supplied by the inventors are:

“1. A computer-implemented method of training and using a machine learning model that controls for consideration of one or more undesired factors which might otherwise be considered by the machine learning model when analyzing new data, the method comprising, via one or more processors: training the machine learning model using a training data set that contains information including the one or more undesired factors; identifying the one or more undesired factors and one or more relevant interaction terms between the one or more undesired factors; and causing the trained machine learning model to not consider the identified one or more undesired factors when analyzing the new data to control for undesired prejudice or discrimination in machine learning models, wherein causing the machine learning model to not consider the identified one or more undesired factors when analyzing the new data includes combining the machine learning model and a second machine learning model to eliminate a bias created by the one or more undesired factors from the machine learning model’s consideration prior to employing the machine learning model to analyze the new data.

“2. The computer-implemented method as set forth in claim 1, wherein the second machine learning model is a linear model.

“3. The computer-implemented method as set forth in claim 1, wherein identifying the one or more undesired factors and the one or more relevant interaction terms between the one or more undesired factors includes training the machine learning model to identify the one or more undesired factors and the one or more relevant interaction terms between the one or more undesired factors.

“4. The computer-implemented method as set forth in claim 3, wherein causing the machine learning model to not consider the identified one or more undesired factors when analyzing the new data includes instructing the machine learning model to not consider the identified one or more undesired factors while analyzing the new data.

“5. The computer-implemented method as set forth in claim 1, wherein the machine learning model is a neural network.

“6. The computer-implemented method as set forth in claim 1, wherein the machine learning model is trained to analyze the new data as part of an underwriting process to determine an appropriate insurance premium.

“7. The computer-implemented method as set forth in claim 6, wherein the new data includes a still image or a video of a person applying for life insurance or health insurance.

“8. The computer-implemented method as set forth in claim 6, wherein the new data includes an image of a piece of property for which a person is applying for property insurance.

“9. The computer-implemented method as set forth in claim 6, wherein the machine learning model is further trained to analyze the new data as part of the underwriting process to determine one or more appropriate terms of coverage.

“10. A computer system configured to train and use a machine learning model that controls for consideration of one or more undesired factors which might otherwise be considered by the machine learning model when analyzing new data, the computer system comprising one or more processors configured to: train the machine learning model using a training data set that contains information including the one or more undesired factors; identify the one or more undesired factors and one or more relevant interaction terms between the one or more undesired factors; and cause the trained machine learning model to not consider the identified one or more undesired factors when analyzing the new data to control for undesired prejudice or discrimination in machine learning models, wherein causing the machine learning model to not consider the identified one or more undesired factors when analyzing the new data includes the one or more processors combining the machine learning model and a second machine learning model to eliminate a bias created by the one or more undesired factors from the machine learning model’s consideration prior to employing the machine learning model to analyze the new data.

“11. The computer system as set forth in claim 10, wherein identifying the one or more undesired factors and the one or more relevant interaction terms between the one or more undesired factors includes the one or more processors training the machine learning model to identify the one or more undesired factors and the one or more relevant interaction terms between the one or more undesired factors.

“12. The computer system as set forth in claim 10, wherein causing the machine learning model to not consider the identified one or more undesired factors when analyzing the new data includes the one or more processors instructing the machine learning model to not consider the identified one or more undesired factors while analyzing the new data.

“13. The computer system as set forth in claim 10, wherein the machine learning model is a neural network.

“14. The computer system as set forth in claim 10, wherein the machine learning model is trained to analyze the new data as part of an underwriting process to determine an appropriate insurance premium.

“15. The computer system as set forth in claim 10, wherein the new data includes images of a person applying for life insurance or health insurance.”

For the URL and additional information on this patent, see: Bernico, Michael L. Method of controlling for undesired factors in machine learning models. U.S. Patent Number 11501133, filed June 4, 2020, and published online on November 15, 2022.

(Our reports deliver fact-based news of research and discoveries from around the world.)