Research map / Reinforcement learning / Offline RL
Adversarially Trained Actor Critic for Offline Reinforcement Learning
Ching-An Cheng, Tengyang Xie, Nan Jiang, Alekh Agarwal
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 ATAC* rather than the strongest listed external baseline | ATAC · D4RL continuous-control tasks | normalized score: 54.3 vs 54.2 on HalfCheetah-medium; 49.5 vs 55.1 on HalfCheetah-medium-replay | Mixed |
| Replace relative pessimism with absolute pessimism (ATAC₀*) (ablation) | ATAC · D4RL locomotion tasks | normalized score: 54.3 -> 43.9 on HalfCheetah-medium; 91.0 -> 90.5 on Walker2d-medium; 102.8 -> 103.5 on Hopper-medium | Mixed |
| Use ATAC₀ rather than ATAC (last iterate) (ablation) | ATAC · D4RL locomotion tasks | normalized score: 53.3 -> 36.8 on HalfCheetah-medium; 89.6 -> 89.6 on Walker2d-medium; 85.6 -> 94.8 on Hopper-medium | Mixed |
| Use intermediate DQRA mixing weight rather than w=1 (double-Q bootstrapping only) (ablation) | ATAC · Hopper-medium-replay | optimization stability | Improved |
| Use intermediate DQRA mixing weight rather than w=0 (residual gradient only) (ablation) | ATAC · Hopper-medium-replay | policy performance | Improved |
| Use w=1, conventional bootstrapping with double Q (ablation) | ATAC · Hopper-medium-replay | optimization stability | Hurt |
| Use w=0, residual gradient only (ablation) | ATAC · Hopper-medium-replay | policy performance | Hurt |
| Use relative rather than absolute pessimism (ablation) | ATAC · D4RL tasks, especially pen-human, door-human, and pen-cloned | normalized score: ATAC₀* 106.1 vs ATAC* 79.3 on pen-human; ATAC₀* 12.2 vs ATAC* 8.7 on door-human; ATAC₀* 104.9 vs ATAC* 73.9 on pen-clone | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- COMBO — COMBO: Conservative Offline Model-Based Policy Optimization (NeurIPS 2021) · better on 3 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Adversarially Trained Weighted Actor-Critic for Safe Offline Reinforcement Learning (NeurIPS 2024 · similarity 0.92)
- Offline-Boosted Actor-Critic: Adaptively Blending Optimal Historical Behaviors in Deep Off-Policy RL (ICML 2024 · similarity 0.90)
- Offline Actor-Critic Reinforcement Learning Scales to Large Models (ICML 2024 · similarity 0.90)
- Off-Policy Actor-Critic for Adversarial Observation Robustness: Virtual Alternative Training via Symmetric Policy Evaluation (ICML 2025 · similarity 0.90)
- Optimistic Critic Reconstruction and Constrained Fine-Tuning for General Offline-to-Online RL (NeurIPS 2024 · similarity 0.89)
- Adversarial Policy Optimization for Offline Preference-based Reinforcement Learning (ICLR 2025 · similarity 0.88)
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.
