Tweeted By @ylecun
Fast similarity search requires quantization.
— Yann LeCun (@ylecun) March 5, 2019
FAIR ICLR paper+code: trains a net to map input distributions to maximally-uniform distributions on the sphere while preserving neighborhood relationships.
Paper: https://t.co/6yBW6iRAEb
Code: https://t.co/U6FxKA2JRw@alexsablay