Research map / Probabilistic and scientific modeling / Variational and generative sampling
Stochastic Beams and Where To Find Them: The Gumbel-Top-k Trick for Sampling Sequences Without Replacement
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use Stochastic Beam Search instead of Diverse Beam Search at the same diversity | Neural machine translation · WMT14 en-fr newstest2014 | mean BLEU | Helped |
| Use Stochastic Beam Search instead of Diverse Beam Search at the same diversity | Neural machine translation · WMT14 en-fr newstest2014 | maximum BLEU | Helped |
| Increase diversity using Stochastic Beam Search rather than Diverse Beam Search at a given BLEU score | Neural machine translation · WMT14 en-fr newstest2014 | n-gram diversity | Helped |
| Use Stochastic Beam Search at low temperatures instead of deterministic Beam Search | Neural machine translation · WMT14 en-fr newstest2014 | maximum BLEU | No clear effect |
| Use Sampling instead of Stochastic Beam Search | Neural machine translation · WMT14 en-fr newstest2014 | mean BLEU | Mixed |
| Use Sampling or Diverse Beam Search instead of Beam Search and Stochastic Beam Search | Neural machine translation · WMT14 en-fr newstest2014 | maximum BLEU | Hurt |
| Normalize the importance weights in the Stochastic Beam Search estimator (ablation) | Stochastic Beam Search BLEU estimator · WMT14 en-fr newstest2014 | estimator variance | Helped |
| Use the normalized Stochastic Beam Search estimator instead of Monte Carlo sampling | Sentence-level BLEU estimation · WMT14 en-fr newstest2014 | estimator variance | Helped |
| Use the normalized Stochastic Beam Search estimator instead of Monte Carlo sampling | Conditional entropy estimation · WMT14 en-fr newstest2014 | estimator variance | Helped |
| Use the normalized Stochastic Beam Search estimator at temperature 0.5 instead of Monte Carlo sampling | Sentence-level BLEU and conditional entropy estimation · WMT14 en-fr newstest2014 | estimator variance | No clear effect |
| Use the normalized rather than the unnormalized priority-sampling estimator (ablation) | Sentence-level BLEU estimation · WMT14 en-fr newstest2014 | estimator bias | Mixed |
Closest papers by meaning
- Discrete Diffusion Models: Novel Analysis and New Sampler Guarantees (NeurIPS 2025 · similarity 0.84)
- Accelerated Speculative Sampling Based on Tree Monte Carlo (ICML 2024 · similarity 0.84)
- Conformal Autoregressive Generation: Beam Search with Coverage Guarantees (AAAI 2024 · similarity 0.84)
- Fast Sampling via Discrete Non-Markov Diffusion Models with Predetermined Transition Time (NeurIPS 2024 · similarity 0.84)
- QUEST: Quality-Aware Metropolis-Hastings Sampling for Machine Translation (NeurIPS 2024 · similarity 0.84)
- Top-H Decoding: Adapting the Creativity and Coherence with Bounded Entropy in Text Generation (NeurIPS 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.