Deep Adaptive Bayesian Screening
Machine Learning
Summary
The authors present a method called Deep Adaptive Bayesian Screening (DABS) to efficiently find important factors in experiments with many possible settings. Their approach learns a strategy ahead of time to choose experiments that give the most useful information, especially when factors can be on or off and may interact. DABS uses a special statistical model to focus on key factors and estimates both how likely factors are active and how big their effects might be. They tested DABS on realistic problems and found it works better and faster than traditional methods when experiments are limited.
Authors
Jade Lejeune Herman, Arno Strouwen, Johan A. K. Suykens, Peter Goos
Abstract
We introduce Deep Adaptive Bayesian Screening (DABS), a method for performing adaptive factorial screening in high-dimensional discrete design spaces. DABS learns a policy network offline to sequentially select informative experiments, amortizing Bayesian Optimal Experimental Design. It handles binary designs, incorporates sparsity and interactions via a spike-and-slab prior with strong heredity. The model is trained using a contrastive lower bound on information about factor activity with nuisance effect sizes and noise variance analytically integrated out. Unlike prior amortized Bayesian design approaches, DABS also integrates Gibbs posterior inference at deployment, yielding posterior probabilities of factor activity and credible intervals on effect sizes. We demonstrate DABS on screening problems calibrated to real-world benchmarks and show it achieves superior accuracy and scalability over classical and Bayesian baselines under tight experimental budgets.