Adversarial Examples, Uncertainty, and Transfer Testing Robustness in Gaussian Process Hybrid Deep Networks

TL;DR

This paper introduces Gaussian Process hybrid deep networks (GPDNNs) that combine DNNs and GPs, resulting in models that are more robust to adversarial attacks and better calibrated in uncertainty estimation, especially under domain shifts.

Contribution

The paper presents GPDNNs, a novel hybrid architecture that inherits the strengths of DNNs and GPs, improving robustness and uncertainty calibration.

Findings

GPDNNs are more robust to adversarial examples.

GPDNNs output high entropy 'don't know' probabilities under domain shift.

GPDNNs outperform pure DNNs in uncertainty estimation.

Abstract

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.