JADAI: Jointly Amortizing Adaptive Design and Bayesian Inference

2025/12/28 by Niels Bracher, Bracher, Niels, Lars Kühmichel +9
Computer Science · #Artificial Intelligence (cs.AI) #Domain Adaptation and Few-Shot Learning #FOS: Computer and information sciences #Generative Adversarial Networks and Image Synthesis #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Machine Learning and Algorithms

paper · doi:10.48550/arxiv.2512.22999

openalex publication_date 2025/12/28 · openalex created_date 2025/12/31 · openalex updated_date 2026/07/28

Abstract

We consider problems of parameter estimation where design variables can be actively optimized to maximize information gain. To this end, we introduce JADAI, a framework that jointly amortizes Bayesian adaptive design and inference by training a policy, a history network, and an inference network end-to-end. The networks minimize a generic loss that aggregates incremental reductions in posterior error along experimental sequences. Inference networks are instantiated with diffusion-based posterior estimators that can approximate high-dimensional and multimodal posteriors at every experimental step. Across standard adaptive design benchmarks, JADAI achieves superior or competitive performance.

Citations

• Diffusion Models in Simulation-Based Inference: A Tutorial Review
• Simulation-Based Inference: A Practical Guide
• Step-DAD: Semi-Amortized Policy-Based Bayesian Experimental Design
• ALINE: Joint Amortization for Bayesian Inference and Active Data Acquisition
• Optimizing Likelihoods via Mutual Information: Bridging Simulation-Based Inference and Bayesian Optimal Experimental Design
• Amortized Bayesian Experimental Design for Decision-Making
• Conditional diffusions for amortized neural posterior estimation
• Amortized Probabilistic Conditioning for Optimization, Simulation and Inference
• Bayesian Experimental Design via Contrastive Diffusions
• Neural Methods for Amortized Inference
• Probabilistic Bayesian optimal experimental design using conditional normalizing flows
• Variational Sequential Optimal Experimental Design using Reinforcement Learning
• Statistically Efficient Bayesian Sequential Experiment Design via Reinforcement Learning with Cross-Entropy Estimators
• Flow Matching for Scalable Simulation-Based Inference
• Understanding Diffusion Objectives as the ELBO with Simple Data Augmentation
• Modern Bayesian Experimental Design
• Simple diffusion: End-to-end diffusion for high resolution images
• Sequential Neural Score Estimation: Likelihood-Free Inference with Conditional Score Based Diffusion Models
• Flow Matching for Generative Modeling
• Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow
• Maximum Likelihood Training for Score-Based Diffusion ODEs by High-Order Denoising Score Matching
• Elucidating the Design Space of Diffusion-Based Generative Models
• Policy-Based Bayesian Experimental Design for Non-Differentiable Implicit Models
• Implicit Deep Adaptive Design: Policy-Based Experimental Design without Likelihoods
• Deep Adaptive Design: Amortizing Sequential Bayesian Experimental Design
• Maximum Likelihood Training of Score-Based Diffusion Models
• Score-Based Generative Modeling through Stochastic Differential Equations
• Neural Approximate Sufficient Statistics for Implicit Models
• Denoising Diffusion Probabilistic Models
• BayesFlow: Learning complex stochastic models with invertible neural networks
• Bayesian Experimental Design for Implicit Models by Mutual Information Neural Estimation
• A Unified Stochastic Gradient Approach to Designing Bayesian-Optimal Experiments
• Optimal experimental design via Bayesian optimization: active causal structure learning for Gaussian process networks
• Generative Modeling by Estimating Gradients of the Data Distribution
• A comprehensive evaluation of full-reference image quality assessment algorithms on KADID-10k
• Variational Bayesian Optimal Experimental Design
• Analyzing Inverse Problems with Invertible Neural Networks
• FiLM: Visual Reasoning with a General Conditioning Layer
• Attention Is All You Need
• An Induced Natural Selection Heuristic for Finding Optimal Bayesian Experimental Designs
• Deep Residual Learning for Image Recognition
• Variational Inference with Normalizing Flows
• On a Measure of the Information Provided by an Experiment

Related

• Deep Adaptive Design: Amortizing Sequential Bayesian Experimental Design
• Implicit Deep Adaptive Design: Policy-Based Experimental Design without Likelihoods
• Bayesian Adaptive Calibration and Optimal Design
• Deep Adaptive Bayesian Screening
• Step-DAD: Semi-Amortized Policy-Based Bayesian Experimental Design