US 12,169,766 B2
Systems and methods for model fairness
Sean Javad Kamkar, Burbank, CA (US); Michael Egan Van Veen, Burbank, CA (US); Feng Li, Burbank, CA (US); Mark Frederick Eberstein, Burbank, CA (US); Jose Efrain Valentin, Burbank, CA (US); Jerome Louis Budzik, Burbank, CA (US); and John Wickens Lamb Merrill, Burbank, CA (US)
Assigned to ZestFinance, Inc., Burbank, CA (US)
Filed by ZestFinance, Inc., Burbank, CA (US)
Filed on Dec. 12, 2023, as Appl. No. 18/536,763.
Application 18/536,763 is a continuation of application No. 17/147,025, filed on Jan. 12, 2021, granted, now 11,893,466.
Application 17/147,025 is a continuation of application No. 16/822,908, filed on Mar. 18, 2020, granted, now 10,977,729, issued on Apr. 13, 2021.
Claims priority of provisional application 62/820,147, filed on Mar. 18, 2019.
Prior Publication US 2024/0127125 A1, Apr. 18, 2024
This patent is subject to a terminal disclaimer.
Int. Cl. G06N 20/20 (2019.01); G06F 18/214 (2023.01); G06N 3/045 (2023.01); G06N 7/01 (2023.01); G06Q 40/03 (2023.01)
CPC G06N 20/20 (2019.01) [G06F 18/2148 (2023.01); G06N 3/045 (2023.01); G06N 7/01 (2023.01); G06Q 40/03 (2023.01)] 22 Claims
OG exemplary drawing
 
1. A method implemented by a modelling system and comprising:
executing a machine learning model building library to iteratively train each of:
an adversarial classifier that is constructed to predict a value for sensitive attributes based on an output generated by a predictive model, wherein iteratively training the adversarial classifier comprises adjusting first parameters of the adversarial classifier to decrease a first value of a first prediction loss metric for the adversarial classifier, and
the predictive model, wherein iteratively training the predictive model comprises adjusting second parameters of the predictive model to decrease a second value of an objective function for the predictive model, wherein the objective function is a difference between a second prediction loss metric for the predictive model and the first prediction loss metric;
determining that a fairness metric threshold is satisfied by the iteratively trained predictive model, wherein predictions of the iteratively trained predictive model provide for improved fairness outcomes as a result of the adversarial classifier correctly classifying one or more of the sensitive attributes; and
providing the iteratively trained predictive model to a model execution system for deployment in a production environment, wherein the iteratively trained predictive model comprises a credit model configured to generate credit decisions with respect to credit applicants.