US 12,333,711 B2
Methods for creating privacy-protecting synthetic data leveraging a constrained generative ensemble model
Engin Dikici, Dublin, OH (US); Luciano Prevedello, Dublin, OH (US); and Matthew Bigelow, Hilliard, OH (US)
Assigned to Ohio State Innovation Foundation, Columbus, OH (US)
Filed by Ohio State Innovation Foundation, Columbus, OH (US)
Filed on Aug. 13, 2021, as Appl. No. 17/401,543.
Claims priority of provisional application 63/065,015, filed on Aug. 13, 2020.
Prior Publication US 2022/0051060 A1, Feb. 17, 2022
Int. Cl. G06N 3/045 (2023.01); G06F 18/21 (2023.01); G06F 18/214 (2023.01); G06F 18/243 (2023.01); G06N 3/04 (2023.01); G06N 3/088 (2023.01); G06N 20/20 (2019.01); G06T 7/00 (2017.01); G06T 11/60 (2006.01); G06V 10/20 (2022.01)
CPC G06T 7/0012 (2013.01) [G06F 18/214 (2023.01); G06F 18/217 (2023.01); G06F 18/243 (2023.01); G06N 3/04 (2013.01); G06N 3/045 (2023.01); G06N 3/088 (2013.01); G06T 11/60 (2013.01); G06V 10/255 (2022.01); G06T 2207/10081 (2013.01); G06T 2207/10088 (2013.01); G06T 2207/10104 (2013.01); G06T 2207/10132 (2013.01); G06T 2207/30096 (2013.01)] 12 Claims
OG exemplary drawing
 
1. A computer-implemented method, comprising:
generating a constrained ensemble of generative adversarial networks (GANs) by:
detecting a poorly-converged ensemble member candidate;
excluding the poorly-converged ensemble member candidate from the constrained ensemble of GANs; and
incrementally adding one or more ensemble member candidates to the constrained ensemble of GANs, wherein the constrained ensemble of GANs comprises a plurality of ensemble members;
analyzing performance of the constrained ensemble of GANs by comparing a temporary performance metric to a baseline performance metric after the one or more ensemble member candidates are added to the constrained ensemble of GANs;
halting generation of the constrained ensemble of GANs to limit membership in the constrained ensemble of GANs based on the comparison;
generating a synthetic dataset using the constrained ensemble of GANs, wherein the synthetic dataset comprises a plurality of synthetic images; and
training a machine learning algorithm using the synthetic images generated by the constrained ensemble of GANs.