Research map / Probabilistic and scientific modeling / Variational and generative sampling
ARMS: Antithetic-REINFORCE-Multi-Sample Gradient for Binary Variables
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 jointly antithetic multi-sample gradients (ARMS) instead of DisARM | Discrete VAE optimizing the ELBO · Dynamic MNIST, Fashion MNIST, and Omniglot | training ELBO: -100.45 -> -98.00 | Helped |
| Use jointly antithetic multi-sample gradients (ARMS) instead of LOORF | Discrete VAE optimizing the ELBO · Dynamic MNIST, Fashion MNIST, and Omniglot | training ELBO: -112.32 -> -112.13 | Helped |
| Use ARMS for the multi-sample bound instead of DisARM | Discrete VAE optimizing the multi-sample bound · Dynamic MNIST, Fashion MNIST, and Omniglot | training multi-sample bound: -100.45 -> -98.00 | Helped |
| Use ARMS for the multi-sample bound instead of VIMCO | Discrete VAE optimizing the multi-sample bound · Dynamic MNIST, Fashion MNIST, and Omniglot | training multi-sample bound: -98.07 -> -98.00 | Helped |
| Use a Dirichlet copula rather than a Gaussian copula (ablation) | ARMS for discrete VAE training · Dynamic MNIST, Fashion MNIST, and Omniglot | training ELBO: -112.13 ± 0.10 vs -111.96 ± 0.09 | No clear effect |
| Use negatively correlated multi-sample copula draws rather than independent samples | ARMS versus LOORF on the toy binary-gradient problem · Toy problem across the tested values of σ(φ) | gradient variance | Helped |
| Increase the number of samples used by pair-based antithetic estimators | DisARM and ARM · Toy binary-gradient problem | gradient variance | Mixed |
Closest papers by meaning
- DisARM: An Antithetic Gradient Estimator for Binary Latent Variables (NeurIPS 2020 · similarity 0.92)
- CARMS: Categorical-Antithetic-REINFORCE Multi-Sample Gradient Estimator (NeurIPS 2021 · similarity 0.90)
- ARM: Augment-REINFORCE-Merge Gradient for Stochastic Binary Networks (ICLR 2019 · similarity 0.88)
- Balancing Performance and Costs in Best Arm Identification (NeurIPS 2025 · similarity 0.83)
- Learning Latent Variable Models via Jarzynski-adjusted Langevin Algorithm (NeurIPS 2025 · similarity 0.83)
- New Classes of the Greedy-Applicable Arm Feature Distributions in the Sparse Linear Bandit Problem (AAAI 2024 · 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.