US Patent Issued to Amazon Technologies on Oct. 27 for "Adversarial training of deep neural networks using non-uniform perturbations" (American, British Inventors)

ALEXANDRIA, Va., Oct. 27 -- United States Patent no. 12,775,836, issued on Oct. 27, was assigned to Amazon Technologies Inc. (Seattle).

"Adversarial training of deep neural networks using non-uniform perturbations" was invented by Sergul Aydore (Brooklyn, N.Y.), Jeffrey Earl Bickford (Thornton, Colo.), Ecenaz Erdemir (London), Luca Melis (New York City, N.Y.) and MohamadAli Torkamani (Scarsdale, N.Y.).

According to the abstract* released by the U.S. Patent & Trademark Office: "Disclosed are systems and methods for training deep neural networks ("DNNs") to increase the detection robustness of those DNNs to adversarial attacks through the use of non-uniform perturbations of training data that consider feature characteristics of features of ...