Research map / Multi-agent systems and games / Human-AI and goal-conditioned
The Neural Testbed: Evaluating Joint Predictions
Ian Osband, Zheng Wen, Seyed Mohammad Asghari, Vikranth Dwaracherla, Xiuyuan Lu, Morteza Ibrahimi, Dieterich Lawson, Botao Hao, Brendan O'Donoghue, Benjamin Van Roy
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 |
|---|---|---|---|
| replace MLP with ensemble+ (adds randomized prior functions) (ablation) | ensemble · Neural Testbed | KL^10: 1.356 -> 1.015 | Improved |
| replace MLP with sgmcmc | MLP · Neural Testbed | KL^10: 1.367 -> 0.947 | Improved |
| replace MLP with hypermodel | MLP · Neural Testbed | KL^10: 1.367 -> 1.107 | Improved |
| replace MLP with ensemble | MLP · Neural Testbed | KL^10: 1.367 -> 1.356 | No clear effect |
| replace MLP with dropout | MLP · Neural Testbed | KL^10: 1.367 -> 1.347 | No clear effect |
| replace MLP with bbb (variational inference) | MLP · Neural Testbed | KL^10: 1.367 -> 1.375 | No clear effect |
| replace MLP with sgmcmc | MLP · Neural Testbed | KL^1 (marginal): 0.129 -> 0.122 | No clear effect |
| replace MLP with ensemble+ | MLP · Neural Testbed | ECE: 0.078 -> 0.085 | Hurt |
| add randomized prior function to ensemble (ensemble+) (ablation) | ensemble · Neural Testbed, low data, medium SNR | KL^10 | Improved |
| add randomized prior function to ensemble (ensemble+) (ablation) | ensemble · Neural Testbed, medium SNR | KL^1 (marginal) | No clear effect |
| increase prediction order tau (ablation) | ensemble+ vs ensemble · Neural Testbed | KL^tau | Improved |
| use agents with better joint prediction (KL^10) in Thompson sampling | Thompson sampling · Neural bandit, N=1000, d=50 | regret at T=50,000 | Improved |
Closest papers by meaning
- Learning Prediction Intervals for Model Performance (AAAI 2021 · similarity 0.84)
- Implicit Neural Representation Inference for Low-Dimensional Bayesian Deep Learning (ICLR 2024 · similarity 0.84)
- From Noise to Diversity: Random Embedding Injection in LLM Reasoning (NeurIPS 2026 · similarity 0.84)
- Test-time Collective Prediction (NeurIPS 2021 · similarity 0.84)
- CoLT: The conditional localization test for assessing the accuracy of neural posterior estimates (NeurIPS 2025 · similarity 0.83)
- SKILL-MIX: a Flexible and Expandable Family of Evaluations for AI Models (ICLR 2024 · similarity 0.83)
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.
