Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Amortized Bayesian Experimental Design with In-Context Knowledge Conditioning
Zhanghu Zhao, Daolang Huang, Xinyi Wen, Samuel Kaski
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes IMBUE, an amortized Bayesian experimental design framework that conditions on in-context knowledge at deployment time so the design policy is not tied to its training prior or task distribution.
Problem
Existing amortized Bayesian experimental design methods are tied to the prior and task distribution used at training and cannot exploit additional information available at deployment.
Approach
Accepts two forms of additional deployment-time information (as stated in the abstract, which is truncated) through a unified knowledge encoding, enabling in-context conditioning of an amortized BED policy rather than retraining per prior.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| concatenate previous instances to ALINE at deployment only (no training exposure) (ablation) | ALINE · CES K=5 | sPCE: 13.37 -> 13.07 (relevant); 13.37 -> 10.16 (irrelevant) | Hurt |
| add mixed-regime relevant/irrelevant training (ALINE-aux) (ablation) | ALINE* · CES K=5 | sPCE: irrelevant 10.16 -> 12.23; relevant 13.07 -> 13.52 | Improved |
| add instance-token interface (ablation) | ALINE-aux · CES K=5 | sPCE: irrelevant 12.23 -> 12.97; relevant 13.52 -> 13.99 | Improved |
| add reliability filter (ablation) | IMBUE w/o filter · CES K=5, irrelevant | sPCE: 12.97 -> 13.14 | Improved |
| add reliability filter (ablation) | IMBUE w/o filter · CES K=5, relevant | sPCE: 13.99 -> 13.93 | No clear effect |
| add reliability filter on prior channel (ablation) | IMBUE · Location Finding | sPCE | No clear effect |
| supply relevant previous instances (K=5) to IMBUE | IMBUE · CES | sPCE: 13.46 -> 13.93 | Improved |
| supply irrelevant previous instances (K=5) to IMBUE | IMBUE · CES | sPCE: 13.46 -> 13.14 | No clear effect |
| use retrieve-then-condition (top two instances) instead of all instances (ablation) | ALINE-aux · CES K=5 | sPCE: irrelevant 12.23 -> 12.69; relevant 13.52 -> 13.29 | Mixed |
| add concatenated previous instances to DAD and RL-BOED (ablation) | DAD, RL-BOED · CES | sPCE: RL-BOED 14.34 -> 2.56 (K=5 relevant) | 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.
- ALINE — ALINE: Joint Amortization for Bayesian Inference and Active Data Acquisition (NeurIPS 2025) · better on 5 of 5 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
- Amortized Bayesian Experimental Design for Decision-Making (NeurIPS 2024 · similarity 0.89)
- Step-DAD: Semi-Amortized Policy-Based Bayesian Experimental Design (ICML 2025 · similarity 0.86)
- ALINE: Joint Amortization for Bayesian Inference and Active Data Acquisition (NeurIPS 2025 · similarity 0.86)
- Bayesian Program Learning by Decompiling Amortized Knowledge (ICML 2024 · similarity 0.85)
- Amortized Active Causal Induction with Deep Reinforcement Learning (NeurIPS 2024 · similarity 0.85)
- Maximin Robust Bayesian Experimental Design (NeurIPS 2026 · similarity 0.84)
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.
