Research map / Reinforcement learning / Visual and representation RL
Causal Influence Detection for Improving Efficiency in Reinforcement Learning
Maximilian Seitzer, Bernhard Schölkopf, Georg Martius
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 |
|---|---|---|---|
| estimate influence with CAI instead of Entropy score | influence detection · Slide | average precision: 0.47 -> 0.98 | Improved |
| estimate influence with CAI instead of Transformer attention weights | influence detection · FetchPickAndPlace | AUC: 0.46 -> 0.97 | Improved |
| estimate influence with CAI instead of simulator Contacts | influence detection · FetchPickAndPlace | AUC: 0.79 -> 0.97 | Improved |
| add CAI reward bonus with appropriate scale to DDPG+HER (ablation) | DDPG+HER · FetchPickAndPlace | episodes to 60% success: reaches 60% success four-times faster | Improved |
| use too large CAI bonus scale (ablation) | DDPG+HER with CAI bonus · FetchPickAndPlace | success rate | Hurt |
| add active causal action exploration on top of epsilon-greedy random exploration (ablation) | DDPG+HER · FetchPickAndPlace | sample efficiency | Improved |
| combine reward bonus with active exploration (ablation) | CAI exploration · FetchPickAndPlace | sample efficiency | Improved |
| use CAI exploration instead of VIME | exploration · FetchPickAndPlace | sample efficiency | Improved |
| use CAI prioritization instead of PER | DDPG+HER · Fetch manipulation tasks (60% success) | learning speed: 1.5-2.5x faster | Improved |
| use CAI prioritization instead of energy-based EBP (privileged) | DDPG+HER · Fetch manipulation tasks | learning speed | No clear effect |
| combine two or three of prioritization, bonus and active exploration (ablation) | DDPG+HER · FetchPickAndPlace | sample efficiency: 4-10x increase | Improved |
| increase observation noise for influence detection (ablation) | CAI · FetchPickAndPlace | average precision | Hurt |
Closest papers by meaning
- Situation-Dependent Causal Influence-Based Cooperative Multi-Agent Reinforcement Learning (AAAI 2024 · similarity 0.89)
- Causal Information Prioritization for Efficient Reinforcement Learning (ICLR 2025 · similarity 0.89)
- Detecting Influence Structures in Multi-Agent Reinforcement Learning (ICML 2024 · similarity 0.88)
- A Snapshot of Influence: A Local Data Attribution Framework for Online Reinforcement Learning (NeurIPS 2025 · similarity 0.87)
- Hierarchical Reinforcement Learning with Targeted Causal Interventions (ICML 2025 · similarity 0.87)
- Causal Action Influence Aware Counterfactual Data Augmentation (ICML 2024 · similarity 0.86)
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.
