Research map / Reinforcement learning / Visual and representation RL
Belief Reward Shaping in Reinforcement Learning
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 |
|---|---|---|---|
| add BRS shaping with optimal parameters to Q-learning | Q-learning · Cliff-jump gridworld | episodes to converge: approximately 700 -> almost immediately | Helped |
| use BRS instead of PBRS/PBA | reward shaping · Cliff-jump gridworld | episodes to converge: approximately 50 -> almost immediately | Helped |
| use small lambda = 100 for BRS (ablation) | BRS · Cliff-jump gridworld | learning speed | Helped |
| use very high prior mean 100 and lambda 10000 (ablation) | BRS · Cliff-jump gridworld | learning speed | Hurt |
| use poorly specified PBRS(1,100) (ablation) | PBRS · Cliff-jump gridworld | learning speed | Hurt |
| use only belief cluster B3 (jump prior) (ablation) | BRS · Cliff-jump gridworld | learning speed | Helped |
| use only belief clusters B1 and B2 (ablation) | BRS · Cliff-jump gridworld | learning speed | Hurt |
| add simple prior beliefs (SPNN) to TD network | TD-Gammon NN · Backgammon vs BL | games to reach level: 17,000 -> 10,000 games | Helped |
| add complex prior beliefs (CPNN) to TD network | TD-Gammon NN · Backgammon vs BL | games to reach 45% points: 24,000 -> 10,000 games | Helped |
| replace decaying belief reward with constant reward equal to prior mean (SCNN) (ablation) | SPNN · Backgammon vs BL | points won share | Hurt |
| replace decaying belief reward with constant reward equal to prior mean (CCNN) (ablation) | CPNN · Backgammon vs BL | points won share | Hurt |
Closest papers by meaning
- BAMDP Shaping: a Unified Framework for Intrinsic Motivation and Reward Shaping (ICLR 2025 · similarity 0.91)
- Reward Shaping for Reinforcement Learning with An Assistant Reward Agent (ICML 2024 · similarity 0.88)
- Highly Efficient Self-Adaptive Reward Shaping for Reinforcement Learning (ICLR 2025 · similarity 0.88)
- Automatic Reward Shaping from Confounded Offline Data (ICML 2025 · similarity 0.88)
- Bootstrapped Reward Shaping (AAAI 2025 · similarity 0.88)
- Long-Horizon Q-Learning: Accurate Value Learning via n-Step Inequalities (NeurIPS 2026 · similarity 0.88)
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.