Research map / Reinforcement learning / Imitation learning
Distributional Inverse Reinforcement Learning
Feiyang Wu, Ye Zhao, Anqi Wu
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 |
|---|---|---|---|
| use distributional reward with FSD loss instead of deterministic reward (Det-TD-Mean) (ablation) | DistIRL · HalfCheetah ablation | scaled performance: 0.22 -> 1.00 | Improved |
| use FSD loss instead of mean matching with distributional reward and quantile critic (ablation) | DistIRL · HalfCheetah ablation | scaled performance: 0.22 -> 1.00 | Improved |
| use distributional reward instead of deterministic reward with quantile critic and mean matching (ablation) | Qt-Mean · HalfCheetah ablation | scaled performance: 0.00 -> 0.22 | Improved |
| use FSD loss instead of mean matching with TD critic (ablation) | Dis-TD · HalfCheetah ablation | scaled performance: 0.33 -> 0.67 | Improved |
| use quantile critic instead of TD critic with FSD loss (ablation) | DistIRL · HalfCheetah ablation | scaled performance: 0.67 -> 1.00 | Improved |
| use distributional reward IRL instead of offline baselines in risk-sensitive setting | DistIRL · Risk-sensitive D4RL HalfCheetah | return: 1259 -> 3469 | Improved |
| use DistIRL instead of BC | DistIRL · Risk-sensitive D4RL Hopper | return: 346 -> 886 | Improved |
| use Gaussian reward instead of quantile-function reward parameterization | DistIRL · Risk-sensitive D4RL | return: 3294 -> 3469 on HalfCheetah | Mixed |
| use DistIRL instead of Offline ML-IRL in risk-neutral setting | DistIRL · Risk-neutral D4RL HalfCheetah | return: 11231 -> 7779 | Hurt |
| use DistIRL instead of Offline ML-IRL in risk-neutral setting | DistIRL · Risk-neutral D4RL Walker2d | return: 4201 -> 4570 | Improved |
| use skew-normal reward model instead of Gaussian | DistIRL · dopamine tracking | correlation / W-1 | Improved |
| change distortion risk measure to one far from expert risk preference (ablation) | DistIRL · D4RL appendix | return | Hurt |
Closest papers by meaning
- Is Inverse Reinforcement Learning Harder than Standard Reinforcement Learning? A Theoretical Perspective (ICML 2024 · similarity 0.90)
- Recent Advancements in Inverse Reinforcement Learning (AAAI 2024 · similarity 0.90)
- Offline Inverse RL: New Solution Concepts and Provably Efficient Algorithms (ICML 2024 · similarity 0.89)
- Rethinking Inverse Reinforcement Learning: from Data Alignment to Task Alignment (NeurIPS 2024 · similarity 0.89)
- Inverse Reinforcement Learning by Estimating Expertise of Demonstrators (AAAI 2025 · similarity 0.89)
- The Distributional Reward Critic Framework for Reinforcement Learning Under Perturbed Rewards (AAAI 2025 · 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.
