We present our work on similarity search. We discuss the metrics of this task, and what families of techniques are appropriate to strike several tradeoffs in the space of those metrics. Then we present several recent improvements in this domain, with a particular focus on a machine learning technique that improves the conditioning of the problem. We also introduce the Faiss library that bundles together many of those techniques.
试看 5 分钟 ，登录 后观看 01:23:33 的高清完整视频