Direct Bethe Free Energy Minimization for Bayesian Neural Networks
基于贝叶斯神经网络的直接贝叶斯自由能最小化
机构 * Cisco Inc.（思科公司）
AI总结 本文提出通过直接最小化贝叶斯自由能来训练贝叶斯神经网络，该方法在树状因子图上精确，可处理混合概率和确定性子图，且在单次梯度传递中实现联合优化，无需交叉验证。
Comments Improved the paper - formalization of routing, add classification experiments, new bib items, free routing in linear model, and many more. Also polished the text
