English

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.

Keywords

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

R2 v1 2026-06-23T19:44:37.084Z