Research map / Reinforcement learning / Visual and representation RL
Exploration by Maximizing Renyi Entropy for Reward-Free RL Framework
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 |
|---|---|---|---|
| maximize state-action Rnyi entropy instead of state entropy with an entropy bonus (ablation) | MaxRenyi · MultiRooms | lowest normalized cumulative reward: State + bonus: 0.14, 0.48, 0.78, 0.87; MaxRenyi: 0.14, 0.52, 0.87, 0.91 | Helped |
| add an entropy bonus to state-entropy exploration (ablation) | State · MultiRooms | lowest normalized cumulative reward: State: 0.07, 0.15, 0.41, 0.57; State + bonus: 0.14, 0.48, 0.78, 0.87 | Helped |
| use PPO with a value function instead of vanilla policy gradient (ablation) | MaxRenyi · MultiRooms exploration | number of unique state-action pairs visited | Helped |
| use alpha=0.5 rather than alpha=1 (ablation) | MaxRenyi · MultiRooms exploration | number of unique state-action pairs visited | Helped |
| use alpha=0 rather than alpha=0.5 (ablation) | MaxRenyi · MultiRooms exploration | number of unique state-action pairs visited | Hurt |
| use MaxRenyi exploration rather than RND, ICM, or MaxEnt | reward-free RL · MultiRooms (planning phase) | sample count needed for normalized cumulative reward larger than 0.8: 8k samples; 16k samples | Helped |
| use MaxRenyi exploration rather than RND | reward-free RL · Gravitar | planned policy score: 1,850 vs 1,890 | Hurt |
| use MaxRenyi exploration rather than RND | reward-free RL · Atari and Mujoco planning tasks | planned policy score: Montezuma 8,100 vs 4,740; Venture 900 vs 840; Gravitar 1,850 vs 1,890; PrivateEye 6,820 vs 3,040; Solaris 2,192 vs 1,416 | Mixed |
Closest papers by meaning
- Behavioral Entropy-Guided Dataset Generation for Offline Reinforcement Learning (ICLR 2025 · similarity 0.89)
- MaxInfoRL: Boosting exploration in reinforcement learning through information gain maximization (ICLR 2025 · similarity 0.89)
- Reward-Free Kernel-Based Reinforcement Learning (ICML 2024 · similarity 0.89)
- Beyond Optimism: Exploration With Partially Observable Rewards (NeurIPS 2024 · similarity 0.89)
- Exploration-Driven Policy Optimization in RLHF: Theoretical Insights on Efficient Data Utilization (ICML 2024 · similarity 0.89)
- DIME: Diffusion-Based Maximum Entropy Reinforcement Learning (ICML 2025 · similarity 0.89)
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.