Research map / Reinforcement learning / Policy gradient and actor-critic
Distributional Reinforcement Learning via Moment Matching
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 |
|---|---|---|---|
| Replace QR-DQN-1's predefined quantile statistics with MMD-learned unrestricted particles | DQN-style distributional RL · Atari-55, 30-no-op evaluation | mean best human-normalized score: 902% -> 1969% | Helped |
| Replace C51's predefined categorical statistics with MMD-learned unrestricted particles | DQN-style distributional RL · Atari-55, 30-no-op evaluation | mean best human-normalized score: 701% -> 1969% | Helped |
| Use MMD-learned unrestricted particles instead of QR-DQN-1 statistics | DQN-style distributional RL · Atari-55, 30-noop evaluation | games above DQN: 54 -> 55 | Helped |
| Use Gaussian kernels with a mixture of bandwidths instead of individual-bandwidth Gaussian or unrectified kernels (ablation) | MMDQN · Six Atari tuning games | human-normalized score | Helped |
| Increase particle count; too small a particle count adversely affects performance (ablation) | MMDQN · Six Atari tuning games | human-normalized score | Helped |
| Use Gaussian kernels instead of unrectified kernels (ablation) | MMDRL · Chain environment, return-distribution approximation | higher-order central moments | Helped |
| Use Gaussian kernels instead of the exp-prod kernel (ablation) | MMDRL · Chain environment and Atari games | empirical performance | Helped |
| Use Gaussian kernels for MMD matching (ablation) | MMDRL · Distributional Bellman updates | contraction guarantee | Hurt |
Closest papers by meaning
- Statistics and Samples in Distributional Reinforcement Learning (ICML 2019 · similarity 0.93)
- Near-Minimax-Optimal Distributional Reinforcement Learning with a Generative Model (NeurIPS 2024 · similarity 0.91)
- Bellman Unbiasedness: Toward Provably Efficient Distributional Reinforcement Learning with General Value Function Approximation (ICML 2025 · similarity 0.90)
- Distributional Bellman Operators over Mean Embeddings (ICML 2024 · similarity 0.89)
- Distributional Reinforcement Learning with Regularized Wasserstein Loss (NeurIPS 2024 · similarity 0.88)
- A Finite Sample Analysis of Distributional TD Learning with Linear Function Approximation (NeurIPS 2025 · 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.