Stochastic Beams and Where To Find Them: The Gumbel-Top-k Trick for Sampling Sequences Without Replacement
| Authors | |
|---|---|
| Publication date | 2019 |
| Journal | Proceedings of Machine Learning Research |
| Event | 36th International Conference on Machine Learning, ICML 2019 |
| Volume | Issue number | 97 |
| Pages (from-to) | 3499-3508 |
| Organisations |
|
| Abstract |
The well-known Gumbel-Max trick for sampling from a categorical distribution can be extended to sample k elements without replacement. We show how to implicitly apply this ’Gumbel-Top-k’
trick on a factorized distribution over sequences, allowing to draw
exact samples without replacement using a Stochastic Beam Search. Even
for exponentially large domains, the number of model evaluations grows
only linear in k
and the maximum sampled sequence length. The algorithm creates a
theoretical connection between sampling and (deterministic) beam search
and can be used as a principled intermediate alternative. In a
translation task, the proposed method compares favourably against
alternatives to obtain diverse yet good quality translations. We show
that sequences sampled without replacement can be used to construct
low-variance estimators for expected sentence-level BLEU score and model
entropy.
|
| Document type | Article |
| Note | 36th International Conference on Machine Learning (ICML 2019) : Long Beach, California, USA, 9-15 June 2019. . - With supplementary file. - In print proceedings pp. 6133-6145. |
| Language | English |
| Published at | http://proceedings.mlr.press/v97/kool19a.html |
| Other links | https://github.com/wouterkool/stochastic-beam-search http://www.proceedings.com/48979.html |
| Downloads |
kool19a
(Final published version)
|
| Supplementary materials | |
| Permalink to this page | |
