Research map / Reinforcement learning / Policy gradient and actor-critic
DualDICE: Behavior-Agnostic Estimation of Discounted Stationary Distribution Corrections
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 DualDICE rather than TD for stationary-distribution correction estimation | OPE with neural-network function approximation · Cartpole and Reacher | OPE estimate accuracy and stability | Helped |
| Use DualDICE rather than TD | Exact tabular correction estimation · Taxi | log RMSE | No clear effect |
| Use DualDICE rather than TD or IS | OPE · Continuous grid | OPE estimation error | Helped |
| Use DualDICE rather than the evaluated baselines | OPE · Acrobot and Pendulum | OPE performance | Helped |
| Set the convex penalty to f(x)=2/3|x|^(3/2) rather than the quadratic choice (ablation) | DualDICE · Continuous grid | OPE estimation error | Helped |
| Estimate the behavior policy for TD instead of assuming it is known (ablation) | TD correction estimation · Cartpole and Reacher | OPE estimate stability | Hurt |
| Increase behavior-policy similarity to the target policy | DualDICE OPE · Cartpole and Reacher | OPE accuracy | Helped |
Closest papers by meaning
- Relaxed Stationary Distribution Correction Estimation for Improved Offline Policy Optimization (AAAI 2024 · similarity 0.89)
- ODICE: Revealing the Mystery of Distribution Correction Estimation via Orthogonal-gradient Update (ICLR 2024 · similarity 0.89)
- Efficient Policy Evaluation with Offline Data Informed Behavior Policy Design (ICML 2024 · similarity 0.88)
- Iteratively Refined Behavior Regularization for Offline Reinforcement Learning (NeurIPS 2024 · similarity 0.88)
- Doubly Optimal Policy Evaluation for Reinforcement Learning (ICLR 2025 · similarity 0.88)
- Demystifying the Paradox of Importance Sampling with an Estimated History-Dependent Behavior Policy in Off-Policy Evaluation (ICML 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.