Research map / Probabilistic and scientific modeling / Variational and generative sampling
Estimating Gradients for Discrete Random Variables by Sampling 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 the unordered set estimator with a built-in baseline instead of the compared estimators | Categorical VAE, Small (10^2), k = 4 · Categorical VAE, Small (10^2) | gradient log-variance: 6.29 vs 6.29 | Mixed |
| Use the unordered set estimator with a built-in baseline instead of the compared estimators | Categorical VAE, Large (10^20), k = 4 · Categorical VAE, Large (10^20) | gradient log-variance: 7.05 vs 7.06 | Helped |
| Use unordered-set sampling without replacement | Bernoulli toy gradient estimation · Bernoulli toy, high and low entropy | gradient variance | Helped |
| Use the unordered set estimator instead of REINFORCE with replacement, stochastic sum-and-sample, or greedy rollout baseline | TSP with 20 nodes · TSP validation set | optimality gap | Helped |
| Use the unordered set estimator instead of the biased risk estimator | TSP with 20 nodes · TSP validation set | optimality gap | No clear effect |
| Use the unordered set estimator instead of REINFORCE with replacement and its built-in baseline | Categorical VAE ELBO optimization · MNIST binarized by a threshold of 0.5 | -ELBO | No clear effect |
Closest papers by meaning
- Gradient Estimation with Discrete Stein Operators (NeurIPS 2022 · similarity 0.89)
- Rao-Blackwellizing the Straight-Through Gumbel-Softmax Gradient Estimator (ICLR 2021 · similarity 0.89)
- A Temporal-Difference Approach to Policy Gradient Estimation (ICML 2022 · similarity 0.88)
- Coupled Gradient Estimators for Discrete Latent Variables (NeurIPS 2021 · similarity 0.87)
- VarGrad: A Low-Variance Gradient Estimator for Variational Inference (NeurIPS 2020 · similarity 0.87)
- Rao-Blackwellised Reparameterisation Gradients (NeurIPS 2025 · similarity 0.85)
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.