Research map / Reinforcement learning / Policy gradient and actor-critic
Mirror Descent Actor Critic via Bounded Advantage Learning
Ryo Iwaki
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| bound the current-state log-policy term in the critic loss (replace identity with tanh(x/10)) (ablation) | MDAC · Mujoco (6 environments) | baseline-normalized IQM score | Improved |
| bound the successor-state log-policy term in addition to the current one (ablation) | MDAC · Mujoco (6 environments) | baseline-normalized IQM score | Improved |
| bound only the successor term while leaving current term as identity (ablation) | MDAC · Mujoco (6 environments) | baseline-normalized IQM score | Hurt |
| use slower-saturating clip(x/10,-1,1) or tanh(x/10) instead of clip(x,-1,1) or tanh(x) (ablation) | MDAC · Mujoco (6 environments) | baseline-normalized IQM score | Improved |
| use tanh instead of clip as bounding function (ablation) | MDAC · Mujoco (6 environments) | baseline-normalized IQM score | No clear effect |
| use sign(x) as bounding function instead of clip or tanh (ablation) | MDAC · Mujoco (6 environments) | baseline-normalized IQM score | Hurt |
| use time-dependent successor bounding that approaches identity (ablation) | MDAC · Mujoco (6 environments) | baseline-normalized IQM score | Improved |
| vary T1 of time-dependent bounding over 1e5 to 1e6 (ablation) | MDAC · Mujoco (6 environments) | baseline-normalized IQM score | No clear effect |
| use MDAC instead of TD3, SAC and X-SAC | MDAC · Mujoco (6 environments) | baseline-normalized IQM score | Improved |
| use X-SAC instead of SAC | X-SAC · Mujoco (6 environments) | baseline-normalized IQM score | Hurt |
| use MDAC instead of SAC | MDAC · Adroit and DMC dog | aggregated normalized IQM score | Improved |
| use MDAC instead of SAC | MDAC · AdroitHandDoor-v1 | return | No clear effect |
Closest papers by meaning
- Mirror descent actor-critic methods for entropy regularised MDPs in general spaces: stability and convergence (NeurIPS 2026 · similarity 0.92)
- Mirror Descent Beyond Euclidean Stability: An Exponential Separation in Initialization Sensitivity (NeurIPS 2026 · similarity 0.89)
- ReLU to the Rescue: Improve Your On-Policy Actor-Critic with Positive Advantages (ICML 2024 · similarity 0.88)
- A Sharper Global Convergence Analysis for Average Reward Reinforcement Learning via an Actor-Critic Approach (ICML 2025 · similarity 0.87)
- Learning mirror maps in policy mirror descent (ICLR 2025 · similarity 0.87)
- Operator World Models for Reinforcement Learning (NeurIPS 2024 · similarity 0.87)
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.
