Research map / Reinforcement learning / Policy gradient and actor-critic
Managing Temporal Resolution in Continuous Value Estimation: A Fundamental Trade-off
Zichen Zhang, Johannes Kirschner, Junxi Zhang, Francesco Zanini, Alex Ayoub, Masood Dehghan, Dale Schuurmans
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 |
|---|---|---|---|
| vary time step-size h from small to large at fixed data budget | Monte-Carlo policy evaluation · Langevin / 1-D LQR (T=8, a=-1) | MSE of value estimate | Mixed |
| increase data budget B | Monte-Carlo policy evaluation · 1-D LQR | MSE | Improved |
| increase data budget B (optimal step-size) | optimal step-size h* · 1-D LQR | h* | Improved |
| increase absolute drift coefficient a | optimal step-size h* · 1-D LQR | h* | Improved |
| change eigenvalues of A in 3x3 stable random matrices | optimal step-size h* · 3-D LQR finite and infinite horizon | h* | No clear effect |
| use a fixed pre-set step-size regardless of data budget | Monte-Carlo policy evaluation · LQR and MuJoCo | MSE | Hurt |
| tune step-size h in nonlinear control systems | Monte-Carlo policy evaluation · Pendulum, BipedalWalker, six MuJoCo tasks | MSE | Improved |
| increase data budget B in nonlinear systems | optimal step-size h* · MuJoCo and Gym tasks | h* | Mixed |
| set h* by the scaling h* = c_F B^(-1/3) fitted from a small budget | step-size selection · nonlinear environments | h* vs B | Mixed |
| use infinite-horizon scaling h* ~ c_I B^(-1/5) rather than finite-horizon B^(-1/3) | step-size selection · discounted infinite-horizon LQR | h* vs B | Mixed |
Closest papers by meaning
- The surprising efficiency of temporal difference learning for rare event prediction (NeurIPS 2024 · similarity 0.87)
- Mitigating Partial Observability in Sequential Decision Processes via the Lambda Discrepancy (NeurIPS 2024 · similarity 0.87)
- Pausing Policy Learning in Non-stationary Reinforcement Learning (ICML 2024 · similarity 0.87)
- Enhancing Value Function Estimation through First-Order State-Action Dynamics in Offline Reinforcement Learning (ICML 2024 · similarity 0.87)
- Continuous Q-Score Matching: Diffusion Guided Reinforcement Learning for Continuous-Time Control (NeurIPS 2025 · similarity 0.87)
- Discerning Temporal Difference Learning (AAAI 2024 · similarity 0.86)
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.
