Research map / Reinforcement learning / Visual and representation RL
Implicit Deep Adaptive Design: Policy-Based Experimental Design without Likelihoods
Desi R. Ivanova, Adam Foster, Steven Kleinegesse, Michael U. Gutmann, Tom Rainforth
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 |
|---|---|---|---|
| Use iDAD (InfoNCE) instead of SG-BOED | adaptive design for the SIR Model · SIR Model | MI lower bound: 3.752 -> 3.915 | Improved |
| Use iDAD (InfoNCE) instead of the variational myopic approach | adaptive design for Location Finding · Location Finding (4D) | total EIG lower bound: 4.639 ± 0.144 -> 7.750 ± 0.039 | Improved |
| Use iDAD (InfoNCE) instead of likelihood-based DAD | adaptive design for Location Finding · Location Finding (6D) | total EIG lower bound: 6.300 ± 0.030 -> 5.986 ± 0.037 | Hurt |
| Use iDAD (InfoNCE) instead of likelihood-based DAD | adaptive design for Location Finding · Location Finding (4D, 6D, 10D, 20D) | total EIG lower bound: 7.967 ± 0.034 -> 7.750 ± 0.039; 6.300 ± 0.030 -> 5.986 ± 0.037; 3.337 ± 0.039 -> 3.251 ± 0.039; 0.937 ± 0.022 -> 0.871 ± | Mixed |
| Use iDAD (InfoNCE) instead of iDAD (NWJ) | adaptive design for Location Finding · Location Finding (4D, 6D, 10D, 20D) | total EIG lower bound: 7.694 ± 0.045 -> 7.750 ± 0.039; 5.765 ± 0.036 -> 5.986 ± 0.037; 3.252 ± 0.039 -> 3.251 ± 0.039; 0.877 ± 0.022 -> 0.871 ± | Mixed |
| Use iDAD (InfoNCE) rather than iDAD (NWJ) | adaptive design for the pharmacokinetic model · Pharmacokinetic model | total EIG | No clear effect |
| Use an attention-based design history encoder instead of pooling (ablation) | iDAD design network · Location Finding | total EIG lower bound: 7.398 ± 0.040 -> 7.750 ± 0.039 | Improved |
| Use an attention-based critic history encoder instead of pooling (ablation) | iDAD critic network · Location Finding | total EIG lower bound: 7.567 ± 0.037 -> 7.750 ± 0.039 | Improved |
| Use attention in both design and critic history encoders instead of pooling in both (ablation) | iDAD · Location Finding | total EIG lower bound: 7.135 ± 0.034 -> 7.750 ± 0.039 | Improved |
Closest papers by meaning
- Step-DAD: Semi-Amortized Policy-Based Bayesian Experimental Design (ICML 2025 · similarity 0.89)
- PASOA- PArticle baSed Bayesian Optimal Adaptive design (ICML 2024 · similarity 0.85)
- Adaptive Experimental Design to Accelerate Scientific Discovery and Engineering Design (AAAI 2025 · similarity 0.84)
- Amortized Bayesian Experimental Design for Decision-Making (NeurIPS 2024 · similarity 0.84)
- Bayesian Experimental Design Via Contrastive Diffusions (ICLR 2025 · similarity 0.84)
- EngiBench: A Framework for Data-Driven Engineering Design Research (NeurIPS 2025 · 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.
