Research map / Reinforcement learning / Visual and representation RL
Efficient Symbolic Policy Learning with Differentiable Symbolic Expression
Jiaming Guo, Rui Zhang, Shaohui Peng, Qi Yi, Xing Hu, Ruizhi Chen, Zidong Du, Xishan Zhang, Ling Li, Qi Guo, Yunji Chen
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 |
|---|---|---|---|
| replace path selector with L1 norm minimization (ablation) | ESPL · BipedalWalker | average episode reward: 309.43 -> 6.81 | Hurt |
| replace path selector with L1 norm minimization (ablation) | ESPL · Hopper | average episode reward: 2442.48 -> 1546.35 | Hurt |
| replace symbolic network with plain structure (ablation) | ESPL · BipedalWalker | average episode reward: 309.43 -> -11.63 | Hurt |
| replace symbolic network with plain structure (ablation) | ESPL · Hopper | average episode reward: 2442.48 -> 2003.24 | Hurt |
| replace symbolic network with densely connected structure (no arranged operators) (ablation) | ESPL · Hopper | average episode reward: 2442.48 -> 2316.54 | Hurt |
| replace symbolic network with densely connected structure (no arranged operators) (ablation) | ESPL · Pendulum | average episode reward: -151.72 -> -144.08 | Improved |
| use ESPL instead of symbolic regression from pretrained policy (ablation) | regression baseline · CartPole | average episode reward: 211.82 -> 1000 | Improved |
| use ESPL instead of DSP (ablation) | DSP · single-task RL | average rank (lower is better): 4.625 -> 3.5 | Improved |
| use ESPL instead of DSP (training episodes) (ablation) | DSP · CartPole | environment episodes: 2M -> 500 | Improved |
| use ESPL instead of DSP (policy length) | DSP · single-task RL | average symbolic length: 8.25 -> 12.91 | Hurt |
| use ESPL instead of TD3 | TD3 · Hopper | average episode reward: 2743.9 -> 2442.48 | Hurt |
| use CSP symbolic policy instead of PEARL neural policy | PEARL · meta-RL deployment | FLOPs: 60-45000x | Improved |
Closest papers by meaning
- Discovering symbolic policies with deep reinforcement learning (ICML 2021 · similarity 0.95)
- SDRL: Interpretable and Data-Efficient Deep Reinforcement Learning Leveraging Symbolic Planning (AAAI 2019 · similarity 0.91)
- Symbolic Visual Reinforcement Learning: A Scalable Framework With Object-Level Abstraction and Differentiable Expression Search (TPAMI 2024 · similarity 0.90)
- BlendRL: A Framework for Merging Symbolic and Neural Policy Learning (ICLR 2025 · similarity 0.90)
- Mitigating Information Loss in Tree-Based Reinforcement Learning via Direct Optimization (ICLR 2025 · similarity 0.89)
- End-to-End Neuro-Symbolic Reinforcement Learning with Textual Explanations (ICML 2024 · 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.
