Research map / Reinforcement learning / Policy gradient and actor-critic
Posterior Sampling for Deep Reinforcement Learning
Remo Sasso, Michelangelo Conserva, Paulo Rauber
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 |
|---|---|---|---|
| Use posterior sampling of forward models instead of a fixed current model with epsilon-greedy exploration (ablation) | PSDRL on Atari · Freeway | evaluation episode return | Improved |
| Use posterior sampling instead of epsilon-greedy exploration (ablation) | PSDRL on Atari · Atari benchmark | sample efficiency | Improved |
| Reuse and continually train the value network across sampled models instead of reinitializing it for each planning step (ablation) | PSDRL on Atari · Atari benchmark | sample efficiency | Improved |
| Reinitialize the value network at every planning step and quadruple its training iterations (ablation) | PSDRL · Atari benchmark | computational cost: approximately 60% | Hurt |
| Replace randomized value-function baselines with PSDRL | Atari benchmark · 55 Atari games | human-normalized score | Improved |
| Replace DreamerV2 with PSDRL | Atari benchmark · 55 Atari games | sample efficiency | Mixed |
| Use 64 × 64 image inputs rather than the 84 × 84 inputs used by Bootstrapped DQN with randomized priors | PSDRL · BeamRider | average return | Hurt |
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.
- Successor Uncertainties — Successor Uncertainties: Exploration and Uncertainty in Temporal Difference Learning (NeurIPS 2019) · better on 4 of 4 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
- Model-based Reinforcement Learning for Continuous Control with Posterior Sampling (ICML 2021 · similarity 0.90)
- Posterior Sampling with Delayed Feedback for Reinforcement Learning with Linear Function Approximation (NeurIPS 2023 · similarity 0.89)
- Q-learning with Posterior Sampling (ICLR 2026 · similarity 0.89)
- Sampling-Based Safe Reinforcement Learning (NeurIPS 2026 · similarity 0.88)
- Provable and Practical: Efficient Exploration in Reinforcement Learning via Langevin Monte Carlo (ICLR 2024 · similarity 0.87)
- Adversarial Diffusion for Robust Reinforcement Learning (NeurIPS 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.
