Research map / Reinforcement learning / Policy gradient and actor-critic
Tactical Optimism and Pessimism for Deep Reinforcement Learning
Ted Moskovitz, Jack Parker-Holder, Aldo Pacchiano, Michael Arbel, Michael Jordan
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 |
|---|---|---|---|
| add TOP bandit-based optimism/pessimism switching to TD3 | TD3 · MuJoCo Humanoid | average reward: 5386 -> 5899 | Improved |
| add TOP to TD3 | TD3 · MuJoCo HalfCheetah | average reward: 9566 -> 13144 | Improved |
| add TOP to TD3 | TD3 · MuJoCo Ant | average reward: 4242 -> 6336 | Improved |
| add TOP to RAD | RAD · DMControl Cup, Catch 100k | average reward: 488 -> 919 | Improved |
| add TOP to RAD | RAD · DMControl Walker, Walk 500k | average reward: 771 -> 966 | Improved |
| add bandit adaptation on top of distributional critics (ablation) | QR-TD3 · MuJoCo HalfCheetah | average reward: 11170 -> 13144 | Improved |
| add distributional critics to bandit framework (ablation) | ND TOP-TD3 · MuJoCo Ant | average reward: 6096 -> 6336 | Improved |
| add distributional critics to bandit framework (ablation) | ND TOP-RAD · DMControl Cheetah, Run 500k | average reward: 564 -> 803 | Improved |
| adaptive beta vs fixed Optimistic/Pessimistic (ablation) | TOP · HalfCheetah and Hopper | average reward | No clear effect |
| add TOP to RAD on Finger, Spin 100k | RAD · DMControl Finger, Spin 100k | average reward: 729 -> 832 | Mixed |
| add TOP to TD3 on Reacher | TD3 · MuJoCo Reacher | average reward: -4.22 -> -3.85 | No clear effect |
Closest papers by meaning
- Decoupled Policy Actor-Critic: Bridging Pessimism and Risk Awareness in Reinforcement Learning (AAAI 2025 · similarity 0.90)
- Minimax Optimal Reinforcement Learning with Quasi-Optimism (ICLR 2025 · similarity 0.89)
- A Unified Principle of Pessimism for Offline Reinforcement Learning under Model Mismatch (NeurIPS 2024 · similarity 0.89)
- Optimistic Model Rollouts for Pessimistic Offline Policy Optimization (AAAI 2024 · similarity 0.88)
- Beyond Optimism: Exploration With Partially Observable Rewards (NeurIPS 2024 · similarity 0.88)
- Pessimism Meets Risk: Risk-Sensitive Offline Reinforcement Learning (ICML 2024 · 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.
