We develop ShiftMatch, a new training-data-dependent likelihood for out of distribution (OOD) robustness in Bayesian neural networks (BNNs). ShiftMatch is inspired by the training-data-dependent "EmpCov" priors from Izmailov et al. (2021a) and efficiently matches test-time spatial correlations to those at training time. Critically, ShiftMatch is designed to leave neural network training unchanged, allowing it to use publically available samples from pretrained BNNs. Using pre-trained HMC samples, ShiftMatch gives strong performance improvements on CIFAR-10-C, outperforms EmpCov priors, and is perhaps the first Bayesian method capable of convincingly outperforming plain deep ensembles. ShiftMatch can be integrated with non-Bayesian methods like deep ensembles, where it offers smaller, but still considerable, performance improvements. Overall, Bayesian ShiftMatch gave slightly better accuracy than ensembles with ShiftMatch, though they both had very similar log-likelihoods.
翻译:我们开发了TshiftMatch, 这是在Bayesian神经网络中进行分配(OOD)的新的培训数据可靠的可能性。 ShiftMatch的灵感来自Izmailov等人(2021a)的基于培训数据的“EmpCov”前科(EmpCov),并且有效地将测试时空间关系与培训时间的测试时空间关系相匹配。 关键地说, ShiftMatch的设计使神经网络培训保持不变,使其能够使用预先培训的BNNS的公开样本。 使用预先培训的HMC样本, ShiftMatch对CIFAR- 10- C的性能有很强的改进,优于EmpCov的前科,或许是第一种能够令人信服地超越直深层成形的Bayesian方法。 ShiftMatch可以与非Bayesian方法相结合,比如深层的ensembles, 它提供较小但仍相当的性能改进。 总的来说, Bayesian ShiftMatch的精度比SingMatch的精度略高于Sembles, 尽管它们都有非常相似的原木状。