Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Bayesian Ensemble for Sequential Decision-Making
Rui Liu, Enmin Zhao, Lu Wang, Yu Li, Ming Pang, Changping Peng, Zhangang Lin, Ching Law, Jingping Shao
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 Bayesian index-distribution updating (BEB) | ensemble+ · Neural Testbed (d = 2, 10, 50) | regret: 37.0%, 12.8%, and 42.2% lower regret | Improved |
| add Bayesian index-distribution updating (BEB) | hypermodel · Neural Testbed (d = 2, 10, 50) | regret: 69.8%, 22.8%, and 30.3% lower regret | Improved |
| add Bayesian index-distribution updating (BEB) | ensemble+ and hypermodel · Mushroom | regret: 8.7% and 4.8% lower regret | Improved |
| add Bayesian index-distribution updating (BEB) | hypermodel · Yahoo!R6B | number of clicks: 49,676.8 -> 50,322.1 | Improved |
| add Bayesian index-distribution updating (BEB) | ensemble+ · Yahoo!R6B subset (first 1M events) | number of clicks: 2,185.4 -> 2,255.5 | Improved |
| add BEB with exact conjugate Beta update | ensemble+ · Neural Testbed (d = 50) | elapsed wall time (s): 1165.07 -> 1162.82 | No clear effect |
| add BEB with variational-inference index update | hypermodel · Neural Testbed (d = 50) | elapsed wall time (s): 60.16 -> 84.20 | Hurt |
| scale ensemble size from 25 to 100 for ensemble+(BEB) (ablation) | ensemble+(BEB) · Neural Testbed (d = 50) | relative regret reduction (%): 28.23% -> 47.97% | Improved |
| reduce index update frequency to one-third in hypermodel(BEB) (ablation) | hypermodel(BEB) · Neural Testbed (d = 50, index dim 36) | elapsed wall time (s): 90.39 -> 76.99 | Improved |
| reduce index update frequency to one-third in hypermodel(BEB) (ablation) | hypermodel(BEB) · Neural Testbed (d = 50, index dim 36) | relative regret reduction (%): 16.47% -> 4.37% | Hurt |
| use Bayesian Ensemble DQN instead of Random Ensemble DQN | DQN ensembles · MiniGrid Empty-6x6 | average reward: 0.186 -> 0.248 | Improved |
| use Bayesian Ensemble DQN instead of Ensemble DQN | DQN ensembles · MiniGrid LavaGapS5-6x6 | average reward: 0.178 -> 0.35 | Improved |
Closest papers by meaning
- An Analysis of Ensemble Sampling (NeurIPS 2022 · similarity 0.89)
- Martingale Posterior Neural Networks for Fast Sequential Decision Making (NeurIPS 2025 · similarity 0.87)
- Scalable Exploration via Ensemble++ (NeurIPS 2025 · similarity 0.86)
- Ensemble sampling for linear bandits: small ensembles suffice (NeurIPS 2024 · similarity 0.86)
- Batch Ensemble for Variance Dependent Regret in Stochastic Bandits (AAAI 2025 · similarity 0.86)
- Bayesian Exploration Networks (ICML 2024 · similarity 0.85)
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.
