Research map / Reinforcement learning / Imitation learning
Model-Free IRL Using Maximum Likelihood Estimation
Vinamra Jain, Prashant Doshi, Bikramjit Banerjee
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 Q-averaging model-free Q update instead of model-based transitions | MLIRL · Gridworld no noise | final log-likelihood: −36.50 ± 3.12 -> −29.25 ± 6.59 | Improved |
| use Q-averaging model-free Q update instead of model-based transitions | MLIRL · Gridworld 10% noise | final log-likelihood: −45.05 ± 9.07 -> −40.49 ± 0.04 | Improved |
| use Q-averaging instead of model-based MLIRL with estimated transitions | MLIRL · Freeway merging NGSIM | final log-likelihood: −1927.09 ± 1.1 -> −1891.34 ± 0.79 | Improved |
| replace max in Q-learning with Boltzmann-weighted mean (Q-softmax) instead of averaging (ablation) | Q-averaging MLIRL · Gridworld no noise | final log-likelihood: −29.25 ± 6.59 -> −40.39 ± 4.15 | Hurt |
| replace averaging with Boltzmann-weighted mean (Q-softmax) (ablation) | Q-averaging MLIRL · Gridworld 10% noise | final log-likelihood: −40.49 ± 0.04 -> −57.35 ± 0.30 | Hurt |
| replace averaging with Boltzmann-weighted mean (Q-softmax) (ablation) | Q-averaging MLIRL · Freeway merging NGSIM | final log-likelihood: −1891.34 ± 0.79 -> −1914.56 ± 1.38 | Hurt |
| use Q-softmax instead of model-based MLIRL | MLIRL · Gridworld no noise | final log-likelihood: −36.50 ± 3.12 -> −40.39 ± 4.15 | Hurt |
| use Q-softmax instead of model-based MLIRL | MLIRL · Freeway merging NGSIM | final log-likelihood: −1927.09 ± 1.1 -> −1914.56 ± 1.38 | Improved |
| use Q-averaging instead of REIRL (needs baseline trajectories) | REIRL · Freeway merging NGSIM | final log-likelihood: −1920.48 ± 3.24 -> −1891.34 ± 0.79 | Improved |
Closest papers by meaning
- Recent Advancements in Inverse Reinforcement Learning (AAAI 2024 · similarity 0.93)
- Offline Inverse RL: New Solution Concepts and Provably Efficient Algorithms (ICML 2024 · similarity 0.93)
- Is Inverse Reinforcement Learning Harder than Standard Reinforcement Learning? A Theoretical Perspective (ICML 2024 · similarity 0.91)
- Sub-optimal Experts mitigate Ambiguity in Inverse Reinforcement Learning (NeurIPS 2024 · similarity 0.90)
- Inverse Reinforcement Learning by Estimating Expertise of Demonstrators (AAAI 2025 · similarity 0.90)
- How does Inverse RL Scale to Large State Spaces? A Provably Efficient Approach (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.
