Research map / Probabilistic and scientific modeling / Variational and generative sampling
Ancestral Gumbel-Top-k Sampling for Sampling Without Replacement
Wouter Kool, Herke van Hoof, Max Welling
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 ancestral Gumbel-Top-k sampling instead of rejection or naive ancestral sampling without replacement | sampling without replacement from Bayesian network · Random Bayesian networks (10 Bernoulli variables) | model evaluations: fewest iterations/model evaluations to generate the same number of samples | Improved |
| increase parallel processors m from 1 to 100 (ablation) | ancestral Gumbel-Top-k sampling · Random Bayesian network (c=0.5, k=100) | iterations vs total cost: around 300 iterations (cost 300) -> 10 iterations (cost 1000) | Mixed |
| limited parallelism m = 10 (ablation) | ancestral Gumbel-Top-k sampling · Random Bayesian network (c=0.5, k=100) | iterations vs total cost | Improved |
| select next variable by minimum entropy instead of fixed or random order (ablation) | ancestral Gumbel-Top-k sampling (m=1) · Random Bayesian networks, k=100 | iterations | Improved |
| use Stochastic Beam Search instead of Diverse Beam Search | diverse translation sets · WMT14 En-Fr newstest2014 (k=5,10,20) | mean/maximum BLEU at equal diversity | Improved |
| use Stochastic Beam Search instead of Sampling with replacement | diverse translation sets · WMT14 En-Fr newstest2014 (k=5,10,20) | mean BLEU and diversity | Mixed |
| use normalized SBS estimator instead of unnormalized SBS (ablation) | sentence BLEU estimation · WMT14 En-Fr, three sentences | estimator variance | Improved |
| use normalized SBS estimator instead of Monte Carlo | sentence BLEU estimation · WMT14 En-Fr, three sentences (τ < 0.5) | estimator variance | Improved |
| use normalized SBS estimator instead of Monte Carlo at higher temperature | sentence BLEU estimation · WMT14 En-Fr, τ = 0.5 | estimator variance | No clear effect |
| use normalized SBS estimator instead of Monte Carlo for entropy | conditional entropy estimation · WMT14 En-Fr, τ < 0.5 | estimator variance | Improved |
Closest papers by meaning
- Sample Complexity of Branch-length Estimation by Maximum Likelihood (ICML 2025 · similarity 0.85)
- Likelihood-free inference of phylogenetic tree posterior distributions (NeurIPS 2026 · similarity 0.85)
- An Efficient Maximal Ancestral Graph Listing Algorithm (ICML 2024 · similarity 0.85)
- Deep Generative Models for Phylogenetic Inference with Complex Evolutionary Processes (NeurIPS 2026 · similarity 0.84)
- Chain of Log-Concave Markov Chains (ICLR 2024 · similarity 0.83)
- An Efficient Search-and-Score Algorithm for Ancestral Graphs using Multivariate Information Scores for Complex Non-linear and Categorical Data (ICML 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.
