Research map / Probabilistic and scientific modeling / Variational and generative sampling
ARM: Augment-REINFORCE-Merge Gradient for Stochastic Binary Networks
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 ARM instead of Concrete | Linear discrete VAE · MNIST-static | test negative log-likelihood: 107.2 -> 107.2 | No clear effect |
| Use ARM instead of MuProp | Nonlinear discrete VAE · MNIST-static | test negative log-likelihood: 99.1 -> 98.4 ± 0.3 | Helped |
| Use ARM instead of Concrete | Two-layer discrete VAE · MNIST-static | test negative log-likelihood: 95.6 -> 96.7 ± 0.3 | Hurt |
| Use ARM instead of NVIL | Linear discrete VAE · OMNIGLOT | test negative log-likelihood: 117.6 -> 115.8 ± 0.2 | Helped |
| Use ARM instead of NVIL | Nonlinear discrete VAE · OMNIGLOT | test negative log-likelihood: 116.6 -> 117.6 ± 0.4 | Hurt |
| Use ARM instead of REBAR | Two-layer discrete VAE · OMNIGLOT | test negative log-likelihood: 110.8 -> 109.8 ± 0.3 | Helped |
| Use ARM instead of the strongest listed baseline | Stochastic binary network for conditional density estimation · MNIST conditional density estimation | test negative log-likelihood: 58.9 -> 57.9 ± 0.1 | Helped |
| Merge two correlated expectations using shared random numbers (ARM rather than AR) (ablation) | ARM gradient estimator · Toy Bernoulli optimization | gradient estimation variance | Helped |
| Use ARM rather than REINFORCE | Toy Bernoulli optimization · Toy Bernoulli optimization | gradient estimation variance | Helped |
| Use ARM rather than RELAX | Toy Bernoulli optimization · Toy Bernoulli optimization | gradient estimation variance | Helped |
| Use ARM rather than vanilla REINFORCE | MNIST-static discrete VAE · MNIST-static | test negative log-likelihood: 170.1 -> 107.2 ± 0.1 | Helped |
Closest papers by meaning
- DisARM: An Antithetic Gradient Estimator for Binary Latent Variables (NeurIPS 2020 · similarity 0.89)
- ARMS: Antithetic-REINFORCE-Multi-Sample Gradient for Binary Variables (ICML 2021 · similarity 0.88)
- ARSM: Augment-REINFORCE-Swap-Merge Estimator for Gradient Backpropagation Through Categorical Variables (ICML 2019 · similarity 0.88)
- Gradient Estimation for Binary Latent Variables via Gradient Variance Clipping (AAAI 2023 · similarity 0.88)
- Marginalized Stochastic Natural Gradients for Black-Box Variational Inference (ICML 2021 · similarity 0.87)
- Layerwise Progressive Freezing: A Training Scaffold for Depth-Scalable Binary Networks (NeurIPS 2026 · similarity 0.84)
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.