Research map / Reinforcement learning / Imitation learning
Sample-efficient Adversarial Imitation Learning
Dahuin Jung, Hyungyu Lee, Sungroh Yoon
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 |
|---|---|---|---|
| add forward-dynamics loss L_F alone to GAIL (ablation) | GAIL · Ant-v2 (100 expert pairs) | cumulative reward: 4198.2 -> 3329.6 | Hurt |
| add forward-dynamics loss L_F alone to GAIL (ablation) | GAIL · HalfCheetah-v2 (100 expert pairs) | cumulative reward: 2034.6 -> 2330.3 | No clear effect |
| add state and action corruption losses (L_SC, L_AC) alone to GAIL (ablation) | GAIL · Walker2d-v2 (100 expert pairs) | cumulative reward: 3513.4 -> 1028.6 | Hurt |
| add L_SC and L_AC on top of L_F (full method) (ablation) | L_F-only GAIL · HalfCheetah-v2 (100 expert pairs) | cumulative reward: 2330.3 -> 5415.9 | Improved |
| add L_F on top of L_SC and L_AC (full method) (ablation) | L_SC+L_AC GAIL · Ant-v2 (100 expert pairs) | cumulative reward: 2244.0 -> 4554.7 | Improved |
| use swapping corruption instead of random corruption (ablation) | Proposed method · Ant-v2 (100 expert pairs) | cumulative reward: 4263.7 -> 4554.8 | Improved |
| use SimCLR InfoNCE for L_F instead of BYOL (ablation) | Proposed method · Ant-v2 (50 optimal + 50 non-optimal) | cumulative reward: 752.19 -> 4384.71 | Improved |
| use SimCLR InfoNCE for L_F instead of MSE (ablation) | Proposed method · Ant-v2 (50 optimal + 50 non-optimal) | cumulative reward: 2793.48 -> 4384.71 | Improved |
| add manifold mixup alone to 2IWIL baseline (ablation) | 2IWIL · Ant-v2 (25 optimal + 75 non-optimal) | cumulative reward: 478.4 -> 776.3 | No clear effect |
| add manifold mixup alone to 2IWIL baseline (ablation) | 2IWIL · Walker2d-v2 (25 optimal + 75 non-optimal) | cumulative reward: 1981.8 -> 3436.0 | Improved |
| add the proposed method to 2IWIL (ablation) | 2IWIL · Ant-v2 (25 optimal + 75 non-optimal) | cumulative reward: 478.4 -> 3764.6 | Improved |
| add manifold mixup on top of the proposed method (ablation) | Proposed + 2IWIL · Ant-v2 (25 optimal + 75 non-optimal) | cumulative reward: 3764.6 -> 3966.0 | Improved |
Closest papers by meaning
- Unlabeled Imperfect Demonstrations in Adversarial Imitation Learning (AAAI 2023 · similarity 0.92)
- Deterministic and Discriminative Imitation (D2-Imitation): Revisiting Adversarial Imitation for Sample Efficiency (AAAI 2022 · similarity 0.92)
- Imitation Learning via Kernel Mean Embedding (AAAI 2018 · similarity 0.91)
- Adversarial Imitation Learning via Boosting (ICLR 2024 · similarity 0.90)
- Diffusing States and Matching Scores: A New Framework for Imitation Learning (ICLR 2025 · similarity 0.89)
- Robust Visual Imitation Learning with Inverse Dynamics Representations (AAAI 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.
