Adversarial Examples, Uncertainty, and Transfer Testing Robustness in Gaussian Process Hybrid Deep Networks
This addresses the robustness issue in deep learning for real-world applications, though it is an incremental improvement by hybridizing existing methods.
The paper tackles the problem of deep neural networks being overconfident and not capturing uncertainty well, which reduces robustness to adversarial examples and domain shifts. It shows that Gaussian Process hybrid deep networks (GPDNNs) combine the representational power of DNNs with the calibrated uncertainties of GPs, resulting in improved robustness by outputting high entropy probabilities when uncertain.
Deep neural networks (DNNs) have excellent representative power and are state of the art classifiers on many tasks. However, they often do not capture their own uncertainties well making them less robust in the real world as they overconfidently extrapolate and do not notice domain shift. Gaussian processes (GPs) with RBF kernels on the other hand have better calibrated uncertainties and do not overconfidently extrapolate far from data in their training set. However, GPs have poor representational power and do not perform as well as DNNs on complex domains. In this paper we show that GP hybrid deep networks, GPDNNs, (GPs on top of DNNs and trained end-to-end) inherit the nice properties of both GPs and DNNs and are much more robust to adversarial examples. When extrapolating to adversarial examples and testing in domain shift settings, GPDNNs frequently output high entropy class probabilities corresponding to essentially "don't know". GPDNNs are therefore promising as deep architectures that know when they don't know.