JADAI: Jointly Amortizing Adaptive Design and Bayesian Inference
This addresses the challenge of efficient experimental design in Bayesian inference, though it appears incremental as it builds on existing adaptive design methods.
The paper tackles the problem of actively optimizing design variables for parameter estimation to maximize information gain, introducing JADAI, a framework that jointly amortizes Bayesian adaptive design and inference, achieving superior or competitive performance across standard benchmarks.
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.