Timezone: »

Tackling covariate shift with node-based Bayesian neural networks
Trung Trinh · Markus Heinonen · Luigi Acerbi · Samuel Kaski

Tue Jul 19 03:30 PM -- 05:30 PM (PDT) @ Hall E #712

Bayesian neural networks (BNNs) promise improved generalization under covariate shift by providing principled probabilistic representations of epistemic uncertainty. However, weight-based BNNs often struggle with high computational complexity of large-scale architectures and datasets. Node-based BNNs have recently been introduced as scalable alternatives, which induce epistemic uncertainty by multiplying each hidden node with latent random variables, while learning a point-estimate of the weights. In this paper, we interpret these latent noise variables as implicit representations of simple and domain-agnostic data perturbations during training, producing BNNs that perform well under covariate shift due to input corruptions. We observe that the diversity of the implicit corruptions depends on the entropy of the latent variables, and propose a straightforward approach to increase the entropy of these variables during training. We evaluate the method on out-of-distribution image classification benchmarks, and show improved uncertainty estimation of node-based BNNs under covariate shift due to input perturbations. As a side effect, the method also provides robustness against noisy training labels.

Author Information

Trung Trinh (Aalto University)
Markus Heinonen (Aalto University)
Luigi Acerbi (University of Helsinki)
Luigi Acerbi

Assistant Professor at the Department of Computer Science, University of Helsinki. Member of the *Finnish Center for Artificial Intelligence* FCAI.

Samuel Kaski (Aalto University and University of Manchester)

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors