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

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.

References in corpus (14)

• TensorFlow: Large-Scale Machine Learning on Heterogeneous Distributed Systems
• Explaining and Harnessing Adversarial Examples
• End to End Learning for Self-Driving Cars
• Densely Connected Convolutional Networks
• Transferability in Machine Learning: from Phenomena to Black-Box Attacks using Adversarial Samples
• Delving into Transferable Adversarial Examples and Black-box Attacks
• The Space of Transferable Adversarial Examples
• Technical Report on the CleverHans v2.1.0 Adversarial Examples Library
• Dropout Inference in Bayesian Neural Networks with Alpha-divergences
• Universal adversarial perturbations
• Random Feature Expansions for Deep Gaussian Processes
• Thoughts on Massively Scalable Gaussian Processes
• Understanding Probabilistic Sparse Gaussian Process Approximations
• A Unifying Framework for Gaussian Process Pseudo-Point Approximations using Power Expectation Propagation

Cited by in corpus (51)

• A Review of Uncertainty Quantification in Deep Learning: Techniques, Applications and Challenges
• ZOO: Zeroth Order Optimization based Black-box Attacks to Deep Neural Networks without Training Substitute Models
• Uncertainty Estimation Using a Single Deep Deterministic Neural Network
• Adversarial Examples: Attacks and Defenses for Deep Learning
• Threat of Adversarial Attacks on Deep Learning in Computer Vision: A Survey
• Scaling Limits of Wide Neural Networks with Weight Sharing: Gaussian Process Behavior, Gradient Independence, and Neural Tangent Kernel Derivation
• Practical Deep Learning with Bayesian Principles
• Simple and Principled Uncertainty Estimation with Deterministic Deep Learning via Distance Awareness
• When Gaussian Process Meets Big Data: A Review of Scalable GPs
• Deep Convolutional Networks as shallow Gaussian Processes
• On Feature Collapse and Deep Kernel Learning for Single Forward Pass Uncertainty
• Tensor Programs I: Wide Feedforward or Recurrent Neural Networks of Any Architecture are Gaussian Processes
• A Fine-Grained Spectral Perspective on Neural Networks
• Adversarial Phenomenon in the Eyes of Bayesian Deep Learning
• Understanding the Decision Boundary of Deep Neural Networks: An Empirical Study
• Deep Gaussian Processes with Convolutional Kernels
• Star-Galaxy Image Separation with Computationally Efficient Gaussian Process Classification
• Adversarial Attack for Uncertainty Estimation: Identifying Critical Regions in Neural Networks
• GP-Tree: A Gaussian Process Classifier for Few-Shot Incremental Learning
• Calibrating Deep Convolutional Gaussian Processes
• Approximate Inference Turns Deep Networks into Gaussian Processes
• A Gaussian Process perspective on Convolutional Neural Networks
• Defence against adversarial attacks using classical and quantum-enhanced Boltzmann machines
• Variational Implicit Processes
• Input Validation for Neural Networks via Runtime Local Robustness Verification
• Stationary Activations for Uncertainty Calibration in Deep Learning
• Bayesian Adversarial Spheres: Bayesian Inference and Adversarial Examples in a Noiseless Setting
• Adversarial Robustness Guarantees for Classification with Gaussian Processes
• Interpretable Uncertainty Quantification in AI for HEP
• Attacking and Defending Machine Learning Applications of Public Cloud
• Hierarchical Gaussian Process Priors for Bayesian Neural Network Weights
• Customizing an Adversarial Example Generator with Class-Conditional GANs
• Adversarial Robustness Guarantees for Gaussian Processes
• Calibrated simplex-mapping classification
• Ptolemy: Architecture Support for Robust Deep Learning
• Sparse Uncertainty Representation in Deep Learning with Inducing Weights
• Butterfly Effect: Bidirectional Control of Classification Performance by Small Additive Perturbation
• Uncertainty-Aware Time-to-Event Prediction using Deep Kernel Accelerated Failure Time Models
• A Modulation Layer to Increase Neural Network Robustness Against Data Quality Issues
• Star-Galaxy Separation via Gaussian Processes with Model Reduction
• Last Layer Marginal Likelihood for Invariance Learning
• Kernelized Capsule Networks
• On The Utility of Conditional Generation Based Mutual Information for Characterizing Adversarial Subspaces
• A Study on the Uncertainty of Convolutional Layers in Deep Neural Networks
• Quantifying Predictive Uncertainty in Medical Image Analysis with Deep Kernel Learning
• Stochastic-Shield: A Probabilistic Approach Towards Training-Free Adversarial Defense in Quantized CNNs
• Probabilistic Modeling for Novelty Detection with Applications to Fraud Identification
• Joints in Random Forests
• Deep Minimax Probability Machine
• Deep Bayesian Image Set Classification: A Defence Approach against Adversarial Attacks
• Towards Robust Classification with Deep Generative Forests