Research map / Reinforcement learning / Offline RL
Universal Value-Function Uncertainties
Moritz Akiya Zanger, Max Weltevrede, Yaniv Oren, Pascal R. Van der Vaart, Caroline Horsch, Wendelin Boehmer, Matthijs T. J. Spaan
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 UVU-based task rejection to DQN | DQN · GoToDoor grid size 10 | average return: 3.39 -> 6.72 | Improved |
| add UVU-based task rejection to DQN | DQN · GoToDoor grid size 5 | average return: 5.50 -> 10.54 | Improved |
| use UVU instead of DQN-RND uncertainty | DQN-RND · GoToDoor grid size 10 | average return: 2.25 -> 6.72 | Improved |
| add uncertainty prior to DQN-RND | DQN-RND · GoToDoor grid size 5 | average return: 3.94 -> 10.41 | Improved |
| use UVU instead of BDQNP(35) ensemble | BDQNP(35) · GoToDoor grid size 9 | average return: 7.03 -> 7.29 | Improved |
| use UVU instead of BDQNP(35) ensemble | BDQNP(35) · GoToDoor grid size 5 | average return: 10.58 -> 10.54 | No clear effect |
| use UVU instead of BDQNP(35) ensemble | BDQNP(35) · GoToDoor grid size 6 | average return: 9.57 -> 9.54 | No clear effect |
| increase ensemble size from 15 to 35 | BDQNP · GoToDoor grid size 10 | average return: 6.09 -> 6.53 | Improved |
| increase ensemble size from 3 to 15 | BDQNP · GoToDoor grid size 5 | average return: 8.69 -> 10.50 | Improved |
| vary network width of UVU (finite width) (ablation) | UVU · GoToDoor grid size 10 | average return | No clear effect |
Closest papers by meaning
- Model-Value Inconsistency as a Signal for Epistemic Uncertainty (ICML 2022 · similarity 0.88)
- Why So Pessimistic? Estimating Uncertainties for Offline RL through Ensembles, and Why Their Independence Matters (NeurIPS 2022 · similarity 0.88)
- Deterministic Uncertainty Propagation for Improved Model-Based Offline Reinforcement Learning (NeurIPS 2024 · similarity 0.88)
- Minimax Optimal and Computationally Efficient Algorithms for Distributionally Robust Offline Reinforcement Learning (NeurIPS 2024 · similarity 0.87)
- Enhancing Value Function Estimation through First-Order State-Action Dynamics in Offline Reinforcement Learning (ICML 2024 · similarity 0.87)
- Learning the Target Network in Function Space (ICML 2024 · 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.
