Research map / Reinforcement learning / Visual and representation RL
Symbolic Visual Reinforcement Learning: A Scalable Framework With Object-Level Abstraction and Differentiable Expression Search
Wenqing Zheng, S P Sharan, Zhiwen Fan, Kuanxiang Wang, Yihan Xi, Zhangyang Wang
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 neural-guided one-tree-at-a-time search (ablation) | DiffSES · Pong (ablation) | reward: 6.4 -> 20.2 | Improved |
| add neural-guided search (ablation) | DiffSES · Pong (ablation) | convergence generations: 103 -> 82 | Improved |
| add SGD coefficient tuning to GP (ablation) | DiffSES · Pong (ablation) | reward: 12.2 -> 20.2 | Improved |
| add SGD coefficient tuning to GP (ablation) | DiffSES · Pong (ablation) | time (s): 6275 -> 5807 | Improved |
| train object detector longer (30% to 100% of training) (ablation) | DiffSES · AdventureIsland3-Nes | reward: 1091 -> 3250 | Improved |
| train object detector to 80% instead of 100% (ablation) | DiffSES · AdventureIsland3-Nes | reward: 3250 -> 2681 | Hurt |
| drop crucial objects with probability 0.9 (ablation) | DiffSES · CircusCharlie-Nes | reward: 7690 -> 900 | Hurt |
| drop non-crucial money bag objects (ablation) | DiffSES · CircusCharlie-Nes | reward: 7690 -> 7690 | No clear effect |
| drop crucial objects with probability 0.3 (ablation) | DiffSES · CircusCharlie-Nes | reward: 7690 -> 4200 | Hurt |
| use object-level DiffSES instead of pixel-level CGP (ablation) | DiffSES · Pong | reward: 19.7 -> 20.2 | Improved |
| use object-level DiffSES instead of pixel-level CGP (ablation) | DiffSES · SpaceInvaders | reward: 713.60 -> 792.39 | Improved |
| distill symbolic policy from PPO teacher instead of using the CNN agent (ablation) | DiffSES · AstroRoboSasa-Nes | reward: 966 -> 1256 | Improved |
Closest papers by meaning
- Breaking the SFT Plateau: Multimodal Structured Reinforcement Learning for Chart-to-Code Generation (ICLR 2026 · similarity 0.87)
- VLMs-Guided Representation Distillation for Efficient Vision-Based Reinforcement Learning (CVPR 2025 · similarity 0.87)
- Hierarchical Process Reward Models are Symbolic Vision Learners (CVPR 2026 · similarity 0.87)
- The Surprising Ineffectiveness of Pre-Trained Visual Representations for Model-Based Reinforcement Learning (NeurIPS 2024 · similarity 0.87)
- BlendRL: A Framework for Merging Symbolic and Neural Policy Learning (ICLR 2025 · similarity 0.87)
- When Should We Prefer State-to-Visual DAgger over Visual Reinforcement Learning? (AAAI 2025 · 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.
