Successive Halving Top-k Operator
Machine Learning
2020-10-30 v1
Abstract
We propose a differentiable successive halving method of relaxing the top-k operator, rendering gradient-based optimization possible. The need to perform softmax iteratively on the entire vector of scores is avoided by using a tournament-style selection. As a result, a much better approximation of top-k with lower computational cost is achieved compared to the previous approach.
Cite
@article{arxiv.2010.15552,
title = {Successive Halving Top-k Operator},
author = {Michał Pietruszka and Łukasz Borchmann and Filip Graliński},
journal= {arXiv preprint arXiv:2010.15552},
year = {2020}
}
Comments
Work in progress