Research map / Reinforcement learning / Imitation learning
Of Moments and Matching: A Game-Theoretic Framework for Closing the Imitation Gap
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 |
|---|---|---|---|
| Match reward moments rather than off-policy Q moments | Moment-matching imitation learning · Imitation learning | imitation gap upper bound: O(εT^2) -> O(εT) | Helped |
| Use on-policy Q moment matching in an H-recoverable problem instead of off-policy Q matching | Moment-matching imitation learning · H-recoverable MDPs | imitation gap upper bound: O(εT^2) -> O(εHT) | Helped |
| Apply AdRIL instead of GAIL | Continuous-control tasks · PyBullet suite | policy performance | Helped |
| Apply AdRIL instead of SQIL | Continuous-control tasks · PyBullet suite | policy performance stability | Helped |
| Use AdRIL rather than behavioral cloning with limited data | HalfCheetah · HalfCheetah Bullet Task | data efficiency | Helped |
| Apply AdVIL instead of ValueDICE | Off-policy imitation learning · HalfCheetah and Ant | compute required to converge: an order of magnitude less compute | Helped |
| Apply DAeQuIL instead of DAgger | Toy UAV navigation task · Forest navigation | policy performance | Helped |
| Use balanced expert/learner trajectory sampling instead of standard replay-buffer sampling (ablation) | AdRIL · Ant environment and other tested tasks | update stability and learner performance | Mixed |
| Train AdVIL longer than the prescribed steps (ablation) | AdVIL · HalfCheetah and Ant | policy performance | Hurt |
| Keep the discriminator fixed rather than updating it adaptively (ablation) | Reward moment-matching imitation learning (SQIL) · Imitation learning | solution quality | Hurt |
Closest papers by meaning
- Efficient Imitation under Misspecification (ICLR 2025 · similarity 0.90)
- Causal Imitation for Markov Decision Processes: a Partial Identification Approach (NeurIPS 2024 · similarity 0.90)
- Interactive and Hybrid Imitation Learning: Provably Beating Behavior Cloning (NeurIPS 2025 · similarity 0.89)
- Imitation Learning from Observation with Automatic Discount Scheduling (ICLR 2024 · similarity 0.89)
- Is Behavior Cloning All You Need? Understanding Horizon in Imitation Learning (NeurIPS 2024 · similarity 0.89)
- Multi-Agent Imitation Learning: Value is Easy, Regret is Hard (NeurIPS 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.