US 12,307,740 B2
Techniques to perform global attribution mappings to provide insights in neural networks
Mark Ibrahim, Brooklyn, NY (US); John Paisley, New York, NY (US); Ceena Modarres, Brooklyn, NY (US); and Melissa Louie, Brooklyn, NY (US)
Assigned to Capital One Services, LLC, McLean, VA (US)
Filed by Capital One Services, LLC, McLean, VA (US)
Filed on Mar. 8, 2024, as Appl. No. 18/600,349.
Application 18/600,349 is a continuation of application No. 17/990,050, filed on Nov. 18, 2022, granted, now 11,928,853.
Application 17/990,050 is a continuation of application No. 16/855,685, filed on Apr. 22, 2020, granted, now 11,568,263, issued on Jan. 31, 2023.
Claims priority of provisional application 62/970,652, filed on Feb. 5, 2020.
Prior Publication US 2024/0355091 A1, Oct. 24, 2024
This patent is subject to a terminal disclaimer.
Int. Cl. G06V 10/762 (2022.01); G06F 18/23213 (2023.01); G06N 3/04 (2023.01); G06N 3/084 (2023.01)
CPC G06V 10/763 (2022.01) [G06F 18/23213 (2023.01); G06N 3/04 (2013.01); G06N 3/084 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method, comprising:
accessing, by processing circuit, a set of credit risk assessment data samples from a data store;
generating, by the processing circuit, local attributions for the set of credit risk assessment data samples;
comparing, by the processing circuit, a pair of local attributions, and assigning a rank distance thereto;
generating clusters of the local attributions, each cluster having a medoid;
generating, by the processing circuit, global attributions, wherein each global attribution corresponds to the medoid of one of the clusters; and
determining, by the processing circuit, a bias in a neural network, one or more features to include in the neural network, or a combination thereof based on the global attributions.