A learning framework for nearest neighbor search
Lawrence Cayton & Sanjoy Dasgupta - UCSD

poster M14 (monday)

idea: Use sample queries to learn a data structure for fast NN retrieval.
eg: KD-tree with boundaries optimized over queries .. and a tuned locality sensitive hashing (LSH) structure

we present:

- algorithms to learn these data structures (KD-trees, LSH) - an accompanying generalization theory - an experimental evaulation