{"id":"arxiv/2604.21849","url":"https://arxiv.org/abs/2604.21849","tags":["stat.ML","cs.LG","math.NA","stat.CO"],"score":0,"title":"Beyond Expected Information Gain: Stable Bayesian Optimal Experimental Design with Integral Probability Metrics and Plug-and-Play Extensions","domain":"stat.ML","source":"arxiv","authors":["Di Wu","Ling Liang","Haizhao Yang"],"reviews":0,"subject":"stat.ML","summary":"Bayesian Optimal Experimental Design (BOED) provides a rigorous framework for decision-making tasks in which data acquisition is often the critical bottleneck, especially in resource-constrained settings. Traditionally, BOED typically selects designs by maximizing expected information gain (EIG), commonly defined through the Kullback-Leibler (KL) divergence. However, classical evaluation of EIG often involves challenging nested expectations, and even advanced variational methods leave the underlying log-density-ratio objective unchanged. As a result, support mismatch, tail underestimation,…","category":"research","comments":0,"disputed":false,"division":"stat","mentions":0,"discussed":0,"published":"2026-04-23","reactions":{}}