Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
JADAI: Jointly Amortizing Adaptive Design and Bayesian Inference
Niels Leif Bracher, Lars Kühmichel, Desi R. Ivanova, Xavier Intes, Paul-Christian Bürkner, Stefan T. Radev
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 |
|---|---|---|---|
| detach summary state before the policy to avoid nested BPTT (ablation) | JADAI · Location Finding (T=30, Q30) | sPCE: 7.935 -> 12.708 | Improved |
| sample rollout length r instead of fixing r=30 (ablation) | JADAI · Location Finding (T=30, Q30) | sPCE: 13.186 -> 12.708 | Hurt |
| sample rollout length r instead of fixing r=30 (ablation) | JADAI · Location Finding (intermediate horizon Q10) | sPCE: 6.433 -> 6.783 | Improved |
| use full BPTT instead of truncated window W=5 (ablation) | JADAI · Location Finding (Q30) | sPCE: 12.536 -> 12.708 | Improved |
| use full BPTT instead of truncated window W=5 (ablation) | JADAI · Location Finding (Q10) | sPCE: 6.533 -> 6.783 | Improved |
| add design-prior pretraining before joint training (ablation) | JADAI · Location Finding (T=30) | sPCE: 12.348 -> 12.708 | No clear effect |
| anneal random-design mix-in instead of pretraining (ablation) | JADAI · Location Finding (T=30) | sPCE: 12.352 -> 12.708 | No clear effect |
| train with terminal horizon T=30 instead of T=20 | JADAI · Location Finding (evaluated at T=20) | sPCE: 10.48 -> 10.90 | Improved |
| train with terminal horizon T=30 instead of T=10 | JADAI · CES (evaluated at T=10) | sPCE: 14.76 -> 14.85 | Improved |
| train with terminal horizon T=30 instead of T=10 | JADAI · Location Finding (evaluated at T=10, K=2) | sPCE: 6.47 -> 6.74 | Improved |
| use diffusion model instead of flow matching posterior | JADAI · MNIST Image Discovery (sigma=0) | SSIM/NRMSE: 0.968 -> 0.988 SSIM (FM higher); NRMSE 0.177 vs 0.185 | Mixed |
| use learned policy instead of random design | JADAI posterior network · MNIST Image Discovery (sigma=0) | SSIM: 0.478 -> 0.968 | Improved |
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 2 of 2 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
- Step-DAD: Semi-Amortized Policy-Based Bayesian Experimental Design (ICML 2025 · similarity 0.88)
- Amortized Bayesian Experimental Design for Decision-Making (NeurIPS 2024 · similarity 0.86)
- ALINE: Joint Amortization for Bayesian Inference and Active Data Acquisition (NeurIPS 2025 · similarity 0.86)
- PASOA- PArticle baSed Bayesian Optimal Adaptive design (ICML 2024 · similarity 0.85)
- Efficient Adaptive Data Acquisition via Pretrained Belief Representations (NeurIPS 2026 · 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.
