Research map / Probabilistic and scientific modeling / Variational and generative sampling
CARMS: Categorical-Antithetic-REINFORCE Multi-Sample Gradient Estimator
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 CARMS-I instead of LOORF | nonlinear categorical VAE · Dynamic MNIST, C=3 | training log likelihood: -95.12 ± 0.21 -> -94.85 ± 0.28 | Helped |
| Use CARMS-I instead of LOORF | nonlinear categorical VAE · Fashion MNIST, C=3 | training log likelihood: -233.75 ± 0.10 -> -233.13 ± 0.16 | Helped |
| Use CARMS-I instead of LOORF | nonlinear categorical VAE · Omniglot, C=3 | training log likelihood: -110.33 ± 0.27 -> -110.19 ± 0.23 | Helped |
| Use CARMS-I instead of LOORF | conditional categorical network · Dynamic MNIST, C=3 | training log likelihood: 57.98 ± 0.10 vs 58.19 ± 0.14 | Mixed |
| Use CARMS-I instead of LOORF | conditional categorical network · Fashion MNIST, C=3 | training log likelihood: 132.83 ± 0.08 vs 133.10 ± 0.05 | Mixed |
| Use CARMS-I instead of LOORF | conditional categorical network · Omniglot, C=3 | training log likelihood: 65.57 ± 0.10 vs 65.92 ± 0.26 | Mixed |
| Use the empirical Gumbel-max PMF estimate rather than inverse-CDF sampling (ablation) | CARMS · categorical VAE | log likelihood | Hurt |
| Use antithetic categorical samples with the CARMS debiasing factor rather than independent samples | CARMS · categorical gradient estimation | gradient variance | Helped |
Closest papers by meaning
- ARMS: Antithetic-REINFORCE-Multi-Sample Gradient for Binary Variables (ICML 2021 · similarity 0.90)
- DisARM: An Antithetic Gradient Estimator for Binary Latent Variables (NeurIPS 2020 · similarity 0.88)
- ARSM: Augment-REINFORCE-Swap-Merge Estimator for Gradient Backpropagation Through Categorical Variables (ICML 2019 · similarity 0.87)
- Coupled Gradient Estimators for Discrete Latent Variables (NeurIPS 2021 · similarity 0.87)
- Differentiable Sampling of Categorical Distributions Using the CatLog-Derivative Trick (NeurIPS 2023 · similarity 0.86)
- Training Data Attribution in Diffusion Models via Mirrored Unlearning and Noise-Consistent Skew (NeurIPS 2026 · 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.