Research map / Reinforcement learning / Offline RL
GenDICE: Generalized Offline Estimation of Stationary Values
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 GenDICE instead of the model-based estimator | Offline PageRank on BA, Cora, and Citeseer graphs · Offline PageRank | log KL-divergence | Helped |
| Use GenDICE instead of IS, IPS, or the model-based estimator | Taxi OPE · Taxi | log MSE | Helped |
| Use GenDICE instead of DualDICE | Discounted OPE · Taxi | log MSE | Helped |
| Use GenDICE instead of the discounted OPE baselines | Continuous-control OPE · Reacher | log MSE | Helped |
| Use GenDICE instead of the baselines | Continuous-control OPE · HalfCheetah | log MSE | Helped |
| Use larger learning rates rather than smaller ones (ablation) | GenDICE · OPE with behavior-policy mixture alpha=0.0 | log MSE | Helped |
| Use squared output activation instead of softplus or exponential (ablation) | GenDICE ratio estimator · OPE | log MSE | Helped |
| Increase the discount factor toward 1 (ablation) | GenDICE rather than DualDICE · OPE | log MSE | Helped |
| Remove the ratio constraint penalty (ablation) | GenDICE · Average-reward OPE | log MSE | Hurt |
| Use KL divergence instead of chi-squared divergence (ablation) | GenDICE · Offline PageRank on BA graph | log KL-divergence | Hurt |
| Use alternative divergences, including Wasserstein-1, Jensen-Shannon, Hellinger, or MMD, instead of chi-squared divergence (ablation) | GenDICE · Offline PageRank on BA graph | log KL-divergence | Mixed |
| Vary the constraint penalty weight over the tested range (ablation) | GenDICE with chi-squared divergence · Offline PageRank on BA graph | log KL-divergence | No clear effect |
Closest papers by meaning
- Relaxed Stationary Distribution Correction Estimation for Improved Offline Policy Optimization (AAAI 2024 · similarity 0.89)
- Dynamic Uncertainty Estimation for Offline Reinforcement Learning (AAAI 2025 · similarity 0.87)
- ODICE: Revealing the Mystery of Distribution Correction Estimation via Orthogonal-gradient Update (ICLR 2024 · similarity 0.87)
- Forecasting in Offline Reinforcement Learning for Non-stationary Environments (NeurIPS 2025 · similarity 0.87)
- Compositional Conservatism: A Transductive Approach in Offline Reinforcement Learning (ICLR 2024 · similarity 0.87)
- ComaDICE: Offline Cooperative Multi-Agent Reinforcement Learning with Stationary Distribution Shift Regularization (ICLR 2025 · similarity 0.87)
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.