Research map / Reinforcement learning / Imitation learning
Recent Advancements in Inverse Reinforcement Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
An invited talk abstract surveying three of the author's IRL papers on learning the feasible reward set, its sample complexity, and sub-optimal experts.
Problem
Inverse reinforcement learning is ambiguous because many reward functions can explain the same expert behavior, and its theoretical understanding (especially sample complexity) is limited.
Approach
Reframes IRL as learning the feasible reward set (all rewards consistent with expert behavior) rather than selecting a single reward, postponing the ambiguity; extends this with Hausdorff-distance-based PAC analysis, sample complexity lower bounds, and use of multiple sub-optimal experts with known sub-optimality levels to shrink the feasible set.
Key results
- Reports that the paper by Metelli, Lazzati, and Restelli (2023) introduces the first sample complexity lower bound for learning the feasible reward set and shows uniform sampling matches it up to logarithmic factors when a generative model is available.
- Reports that Poiani et al. (2023) show multiple sub-optimal experts with known sub-optimality bounds significantly reduce the size of the feasible reward set, with almost matching sample complexity algorithms.
Limitations
- Text is a short talk abstract with no experiments reported directly.
- IRL remains less theoretically understood than RL, per the author.
Closest papers by meaning
- Sub-optimal Experts mitigate Ambiguity in Inverse Reinforcement Learning (NeurIPS 2024 · similarity 0.96)
- Offline Inverse RL: New Solution Concepts and Provably Efficient Algorithms (ICML 2024 · similarity 0.95)
- Is Inverse Reinforcement Learning Harder than Standard Reinforcement Learning? A Theoretical Perspective (ICML 2024 · similarity 0.95)
- How does Inverse RL Scale to Large State Spaces? A Provably Efficient Approach (NeurIPS 2024 · similarity 0.92)
- Inverse Reinforcement Learning by Estimating Expertise of Demonstrators (AAAI 2025 · similarity 0.92)
- Towards the Transferability of Rewards Recovered via Regularized Inverse Reinforcement Learning (NeurIPS 2024 · similarity 0.92)
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.