At the seminar, we will discuss various methods of nearest neighbors search.
Current state-of-the-art methods evolved quite a bit since the times of KD-tree and R-tree, and even LSH is old news.
We will go through several novel approximate nearest neighbor methods that are worthy of thorough examination.
We will start with the LSH method, for it is based on quite an interesting idea that birthed both sketch and ANN methods.
Another popular approach is product quantization from facebook faiss library that provides an application for quantization theory.
Finally, we will discuss navigable small world graphs, that allow building model of the space following a principle that is similar to skip-lists and that serves as the base for the nmslib library.
Speaker: Igor Kuralenok.
Presentation language: Russian.
Date and time: February 7th, 20:00-21:30.
Location: Times, room 405.
Videos from seminars will be available at http://bit.ly/MLJBSeminars