Research map / Reinforcement learning / Offline RL
Understanding, Predicting and Better Resolving Q-Value Divergence in Offline-RL
Yang Yue, Rui Lu, Bingyi Kang, Shiji Song, Gao Huang
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 LayerNorm to critic of Diff-QL | Diff-QL · D4RL Antmaze average | normalized score: 29.6 -> 80.8 | Improved |
| add LayerNorm to critic of Diff-QL (best-score metric) | Diff-QL · D4RL Antmaze average | best normalized score: 75.4 -> 87.4 | Improved |
| add LayerNorm to critic | CQL · 1% Mujoco locomotion dataset, 9 tasks | total score: 50 -> 300 | Improved |
| add LayerNorm to critic | TD3+BC · 1% Mujoco locomotion dataset, 9 tasks | total score: 70 -> 228 | Improved |
| add LayerNorm to critic when data is plentiful (100% dataset) | offline RL baselines · D4RL Mujoco locomotion 100% | score | No clear effect |
| weak BC coefficient 0.5 in Diff-QL (ablation) | Diff-QL · Antmaze-large-play | score | Hurt |
| strong BC coefficient 10 in Diff-QL (ablation) | Diff-QL · Antmaze-large-play | score | Hurt |
| add LayerNorm with weak BC 0.5 (ablation) | Diff-QL · Antmaze-large-play | score | Improved |
| replace LayerNorm with WeightNorm (ablation) | TD3 without EMA · walker2d-medium-expert-v2 | SEEM and Q-value stability | Improved |
| add weight decay or dropout (ablation) | TD3 without EMA · walker2d-medium-expert-v2 | SEEM and Q-value stability | No clear effect |
| add BatchNorm to critic (ablation) | TD3 without EMA · walker2d-medium-expert-v2 | SEEM and Q-value stability | Mixed |
| add DR3 regularizer (ablation) | Diff-QL · Antmaze | stability and score | No clear effect |
Closest papers by meaning
- A Perspective of Q-value Estimation on Offline-to-Online Reinforcement Learning (AAAI 2024 · similarity 0.92)
- Long-Horizon Q-Learning: Accurate Value Learning via n-Step Inequalities (NeurIPS 2026 · similarity 0.89)
- Q-Distribution guided Q-learning for offline reinforcement learning: Uncertainty penalized Q-value via consistency model (NeurIPS 2024 · similarity 0.89)
- Model-based Offline Reinforcement Learning with Lower Expectile Q-Learning (ICLR 2025 · similarity 0.89)
- Is Value Learning Really the Main Bottleneck in Offline RL? (NeurIPS 2024 · similarity 0.89)
- Exclusively Penalized Q-learning for Offline Reinforcement Learning (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.
