GSI Takes One of the Top Spots in the Billion-Scale Approximate Nearest Neighbor Search (ANNS) Challenge

About this update from Gsi Technology, Inc.

[{"type":"text","content":"SUNNYVALE, Calif., Dec. 22, 2021 (GLOBE NEWSWIRE) -- GSI Technology, Inc. (Nasdaq: GSIT), a leading provider of high-performance memory solutions for the networking, telecommunications and military markets, and developer of the Gemini® Associative Processing Unit (APU) for Artificial Intelligence, today announced that it was among the leaders in the Billion-Scale Approximate Nearest Neighbor Search (ANNS) Challenge held at this year’s NeurIPS 2021 Conference. Approximate Nearest Neighbor Search is a critical real-world problem facing search, retrieval, and recommendation applications that are used in many different domains. Despite the broad range of algorithms and approaches for ANNS, most empirical evaluations of algorithms have focused on smaller datasets (1 million points). However, deploying recent algorithmic advances in ANNS techniques for search, recommendation and ranking at scale requires technologies that can provide support at a billion, trillion, or larger scale. Currently, there is limited consensus on which algorithms are effective at these larger scales. The Billion-Scale ANNS Challenge was created to provide a comparative understanding of algorithmic ideas and their application at scale, promote the development of new techniques for the problem and demonstrate their value, and introduce a standard benchmarking approach. The six databases used for the challenge were two standard datasets (Deep1B and BigANN), along with two supplied by Microsoft, one by Facebook, and one by Yandex. There were four categories ranked in the challenge: Recall, Thru-put, Power, and Cost. GSI focused on the Recall, or accuracy, category which is the functionality most important in their target applications. For the last several years, the state-of-the-art baseline for large scale ANN has been FAISS (Facebook AI Similarity Search). Many participants withdrew from the challenge because they could not improve beyond the baseline. GSI ranked above the FAISS baseline in all datasets, which is a material accomplishment for a new entrant into the AI sector. Also, GSI was the only participant to attempt the Facebook database. “We are extremely encouraged by our results in the ANNS Challenge as we proved our technology could perform on par with the category leaders in AI,” said Lee-Lean Shu, Chairman, and Chief Executive Officer. “It was an hon...

More updates from Gsi Technology, Inc.