Research map / Probabilistic and scientific modeling / Variational and generative sampling
Gradient Estimation for Binary Latent Variables via Gradient Variance Clipping
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 UGC, switching from DisARM near parameter boundaries to bitflip-1 | Bernoulli gradient estimation · Best subset regression, SNR 0.95 | TPR: 0.40 (0.29) -> 0.67 (0.21) | Helped |
| Use UGC, switching from DisARM near parameter boundaries to bitflip-1 | Bernoulli gradient estimation · Best subset regression, SNR 1.69 | TPR: 0.43 (0.26) -> 0.87 (0.22) | Helped |
| Use UGC instead of bitflip-1 | Bernoulli gradient estimation · Best subset regression, SNR 0.95 | TPR: 0.43 (0.30) -> 0.67 (0.21) | Helped |
| Use UGC instead of DisARM | Bernoulli gradient estimation · P2 toy optimization | convergence to optimum | Helped |
| Use UGC instead of DisARM or bitflip-1 | Discrete VAE training · DynamicMNIST, FashionMNIST, and Omniglot | final optimization objective | Helped |
| Use bitflip-1 or UGC instead of DisARM | Discrete VAE Gaussian mixture model · Gaussian mixture model | gradient variance | Helped |
| Replace UGC with tUGC, selecting the lowest-probability coordinates for bitflip updates (ablation) | UGC · Discrete VAE training | training performance | Helped |
| Use bitflip-1 instead of DisARM | Bernoulli gradient estimation · P1 toy optimization | gradient variance | Mixed |
Closest papers by meaning
- Gradient Estimation with Discrete Stein Operators (NeurIPS 2022 · similarity 0.89)
- REBAR: Low-variance, unbiased gradient estimates for discrete latent variable models (NeurIPS 2017 · similarity 0.89)
- Backpropagation through the Void: Optimizing control variates for black-box gradient estimation (ICLR 2018 · similarity 0.89)
- DisARM: An Antithetic Gradient Estimator for Binary Latent Variables (NeurIPS 2020 · similarity 0.89)
- VarGrad: A Low-Variance Gradient Estimator for Variational Inference (NeurIPS 2020 · similarity 0.88)
- ARM: Augment-REINFORCE-Merge Gradient for Stochastic Binary Networks (ICLR 2019 · similarity 0.88)
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.