Research map / Reinforcement learning / Visual and representation RL
Interpretable and Explainable Logical Policies via Neurally Guided Symbolic Abstraction
Quentin Delfosse, Hikaru Shindo, Devendra Singh Dhami, Kristian Kersting
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 |
|---|---|---|---|
| NUDGE logic policy instead of DQN | Atari agent · Asterix | score: 124.5 -> 6259 | Improved |
| NUDGE logic policy instead of DQN | Atari agent · Freeway | score: 25.8 -> 21.4 | Hurt |
| fix all rule weights to 1.0 (classic logic, no differentiable weights) (ablation) | NUDGE rule set · GetOut | score: 17.86 -> 11.59 | Hurt |
| fix all rule weights to 1.0 (ablation) | NUDGE rule set · Loot | score: 5.66 -> 0.51 | Hurt |
| fix all rule weights to 1.0 (ablation) | NUDGE rule set · 3Fishes | score: 0.05 -> -0.24 | Hurt |
| modify environment to 3Fishes-C by swapping is_bigger atom with same_color, no retraining | NUDGE vs neural PPO · 3Fishes-C | score: -0.37 -> 3.26 | Improved |
| evaluate on larger GetOut+ with 5 enemies without retraining | NUDGE vs neural PPO · GetOut+ | score: -20.88 -> 3.60 | Improved |
| evaluate on Loot-C without retraining | NUDGE vs neural PPO · Loot-C | score: 0.83 -> 5.63 | Improved |
| NUDGE vs neural PPO on harder Loot-Hard | NUDGE · Loot-Hard | score: 8.4 -> 9.2 | No clear effect |
| GetOut+ with 5 enemies vs original GetOut | NUDGE · GetOut+ | score: 17.86 -> 3.60 | Hurt |
| neurally-guided symbolic abstraction vs no-abstraction rule generation (ablation) | NUDGE · GetOut, 3Fishes, Loot | return | Improved |
Closest papers by meaning
- BlendRL: A Framework for Merging Symbolic and Neural Policy Learning (ICLR 2025 · similarity 0.89)
- End-to-End Neuro-Symbolic Reinforcement Learning with Textual Explanations (ICML 2024 · similarity 0.87)
- Mitigating Information Loss in Tree-Based Reinforcement Learning via Direct Optimization (ICLR 2025 · similarity 0.85)
- Neurosymbolic Reinforcement Learning: Playing MiniHack with Probabilistic Logic Shields (AAAI 2025 · similarity 0.85)
- ORCAID: Oblique Rule-Based Continuous-Action Interpretation for Deep RL Policies (NeurIPS 2026 · similarity 0.85)
- LogiCity: Advancing Neuro-Symbolic AI with Abstract Urban Simulation (NeurIPS 2024 · similarity 0.85)
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.
