Research map / Reinforcement learning / Policy gradient and actor-critic
Controlling Overestimation Bias with Truncated Mixture of Continuous Distributional Quantile Critics
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 SAC with TQC, using environment-tuned atom truncation | SAC · Hopper, HalfCheetah, Walker2d, Ant, Humanoid | average seed return: Hopper: $2.86(.58)$ -> $\mathbf{3.71(.16)}$; HalfCheetah: $12.41(5.14)$ -> $\mathbf{18.09(.34)}$; Walker2d: $5.76(.46)$ | Helped |
| Increase SAC critic network size (ablation) | SAC · Humanoid, Walker2d | performance | Mixed |
| Add quantile distributional critics (ablation) | B-SAC · Humanoid, Walker2d | performance | Helped |
| Add individual target-distribution truncation (ablation) | QB-SAC · Humanoid, Walker2d | performance | Mixed |
| Pool already-truncated critic distributions (ablation) | TQB-SAC · Humanoid, Walker2d | performance | Mixed |
| Truncate the pooled critic mixture instead of pooling individually truncated distributions (ablation) | PTQB-SAC · Humanoid, Walker2d | performance | Mixed |
| Add truncation (ablation) | quantile critics · Walker2d, Humanoid | stability and seed variance | Helped |
| Increase the ensemble from two to five Z-networks (ablation) | TQC · Humanoid, Walker2d | performance | Helped |
| Vary the number of dropped atoms per critic (ablation) | TQC · Humanoid | performance | Mixed |
| Increase atoms per critic above 10 while keeping the truncation ratio approximately constant (ablation) | TQC · Humanoid, Walker2d | performance | No clear effect |
| Increase the number of Z-networks from one (ablation) | TQC · Humanoid, Walker2d | performance | Helped |
| Increase the number of Z-networks beyond approximately three (ablation) | TQC · Humanoid, Walker2d | performance | No clear effect |
Closest papers by meaning
- Actor-critic is implicitly biased towards high entropy optimal policies (ICLR 2022 · similarity 0.85)
- Low-Rank Adaptation for Critic Learning in Off-Policy Reinforcement Learning (NeurIPS 2026 · similarity 0.84)
- Overestimation, Overfitting, and Plasticity in Actor-Critic: the Bitter Lesson of Reinforcement Learning (ICML 2024 · similarity 0.84)
- Truncated Gaussian Policy for Debiased Continuous Control (AAAI 2025 · similarity 0.84)
- Quality-Diversity Actor-Critic: Learning High-Performing and Diverse Behaviors via Value and Successor Features Critics (ICML 2024 · similarity 0.84)
- Diffusion Actor-Critic: Formulating Constrained Policy Iteration as Diffusion Noise Regression for Offline Reinforcement Learning (ICLR 2025 · 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.