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

By John Bradshaw, Alexander G. de G. Matthews, Zoubin Ghahramani. 2017

179 citations recorded by the source corpus.

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.

Subfields: Artificial Intelligence, Safety, Robustness & Interpretability, Probabilistic & Bayesian Methods

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Safety, Robustness & Interpretability

In Probabilistic & Bayesian Methods

Citations in the corpus

Simple and Principled Uncertainty Estimation with Deterministic Deep Learning via Distance Awareness (2020, methodology): “Earlier works that combine a GP with a DNN usually perform MAP estimation [11] or structured Variational Inference (VI) ( [9, 78]).”
ZOO: Zeroth Order Optimization Based Black-box Attacks to Deep Neural Networks without Training Substitute Models (2017, background): “For example, the authors in [6] proposed to integrate DNNs with Gaussian processes and RBF kernels for enhanced robustness.”
GP-Tree: A Gaussian Process Classifier for Few-Shot Incremental Learning (2021, methodology): “, 2016b) which learned multiple GPs, each on a different subset of the embedding space, and combined them with the Softmax function; and (2) GPDNN (Bradshaw et al., 2017) which used the Robust-Max likelihood (Hernández-Lobato et al.”
Simple and Scalable Epistemic Uncertainty Estimation Using a Single Deep Deterministic Neural Network (2020, background): “A good visualisation can be found in Bradshaw et al. (2017).”
Approximate Inference Turns Deep Networks into Gaussian Processes (2019, background): “In contrast, GPs are suitable for small datasets and compute confidence estimates, but they are not scalable and choosing a good kernel in practice is challenging [3].”
Scaling Limits of Wide Neural Networks with Weight Sharing: Gaussian Process Behavior, Gradient Independence, and Neural Tangent Kernel Derivation (2019, neutral): “…works have exploited this DNN-GP correspondence implicitly or explicitly to build new models (Cho & Saul, 2009; Lawrence & Moore, 2007; Damianou & Lawrence, 2013; Wilson et al., 2016b;a; Bradshaw et al., 2017; van der Wilk et al., 2017; Kumar et al., 2018; Blomqvist et al., 2018; Borovykh, 2018).”
Practical Deep Learning with Bayesian Principles (2019, background): “Similarly, sequential learning can cause forgetting of past knowledge [27], and lack of reliable confidence estimates and other robustness issues can make it vulnerable to adversarial attacks [6].”
Bayesian Deep Convolutional Networks with Many Channels are Gaussian Processes (2018, background): “Alternatively, deep kernel learning (Wilson et al., 2016b;a; Bradshaw et al., 2017) utilizes GPs with base kernels which take in features produced by a deep neural network (often a CNN), and train the resulting model end-to-end.”
Opportunities and challenges in developing deep learning models using electronic health records data: a systematic review (2018, background): “These include better calibration of uncertainties 116 and adversarial learning with relaxing the shared label space assumption.”
Imitation networks: Few-shot learning of neural networks from scratch (2018, methodology): “ neural networks. In particular for the use of Gaussian processes as referencd models, they introduce local smoothness of predictions into neural networks, which is lacked in standard neural networks [3] and is proved to be effective for overﬁtting [23]. 3.3 Fidelity weighting We note that all the p”
A Fine-Grained Spectral Perspective on Neural Networks (2019, background): “[12] John Bradshaw, Alexander G de G Matthews, and Zoubin Ghahramani.”
Uncertainty Estimation Using a Single Deep Deterministic Neural Network (2020, background): “A good visualisation can be found in Bradshaw et al. (2017).”
Secure and Robust Machine Learning for Healthcare: A Survey (2020, background): “In [169], authors exploited the uncertainty around the”
Deep Convolutional Networks as shallow Gaussian Processes (2018, background): “Wilson et al. [13] introduced and Bradshaw et al. [25] improved deep kernel learning.”