US Patent Issued to GOOGLE on May 5 for "Training embedding models using a stale embedding cache for negative sampling" (New York, New Jersey Inventors)

ALEXANDRIA, Va., May 5 -- United States Patent no. 12,619,922, issued on May 5, was assigned to GOOGLE LLC (Mountain View, Calif.).

"Training embedding models using a stale embedding cache for negative sampling" was invented by Erik Michael Lindgren (New York), Sashank Jakkam Reddi (Jersey City, N.J.), Ruiqi Guo (Elmhurst, N.Y.) and Sanjiv Kumar (Jericho, N.Y.).

According to the abstract* released by the U.S. Patent & Trademark Office: "Provided are systems and methods which more efficiency train embedding models through the use of a cache of item embeddings for candidate items over a number of training iterations. The cached item embeddings can be "stale" embeddings that were generated by a previous version of the model at a previous tra...