用随机集成近似 Bayesian 后验
Bayesian posterior approximation with stochastic ensembles
AI总结：
提出将 dropout 等随机方法与深度集成的随机神经网络集成，通过变分推断近似 Bayesian 后验，在玩具问题和 CIFAR 上比常用基线提供更准确的后验估计。
摘要：
我们引入随机神经网络集成来近似 Bayesian 后验，将 dropout 等随机方法与深度集成相结合。随机集成被表述为分布族，并通过变分推断训练以近似 Bayesian 后验。我们基于 Monte Carlo dropout、DropConnect 以及一种新颖的非参数版 dropout 实现了随机集成，并在一个玩具问题和 CIFAR 图像分类上进行了评估。对于这两项任务，我们直接与 Hamiltonian Monte Carlo 模拟对比测试了后验质量。结果表明，随机集成比其他流行的 Bayesian 推断基线提供了更准确的后验估计。
英文摘要：
We introduce ensembles of stochastic neural networks to approximate the Bayesian posterior, combining stochastic methods such as dropout with deep ensembles. The stochastic ensembles are formulated as families of distributions and trained to approximate the Bayesian posterior with variational inference. We implement stochastic ensembles based on Monte Carlo dropout, DropConnect and a novel non-parametric version of dropout and evaluate them on a toy problem and CIFAR image classification. For both tasks, we test the quality of the posteriors directly against Hamiltonian Monte Carlo simulations. Our results show that stochastic ensembles provide more accurate posterior estimates than other popular baselines for Bayesian inference.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
