Towards Deep Learning Models Resistant to Adversarial Attacks

Published in

International Conference on Learning Representa...(2017)

External Links:

Generate Graph

TL;DR

This work studies the adversarial robustness of neural networks through the lens of robust optimization, and suggests the notion of security against a first-order adversary as a natural and broad security guarantee.

Abstract

Recent work has demonstrated that deep neural networks are vulnerable to adversarial examples---inputs that are almost indistinguishable from natural data and yet classified incorrectly by the network. In fact, some of the latest findings suggest that the existence of adversarial attacks may be an inherent weakness of deep learning models. To address this problem, we study the adversarial robustness of neural networks through the lens of robust optimization. This approach provides us with a broad and unifying view on much of the prior work on this topic. Its principled nature also enables us to identify methods for both training and attacking neural networks that are reliable and, in a certain sense, universal. In particular, they specify a concrete security guarantee that would protect against any adversary. These methods let us train networks with significantly improved resistance to a wide range of adversarial attacks. They also suggest the notion of security against a first-order adversary as a natural and broad security guarantee. We believe that robustness against such well-defined classes of adversaries is an important stepping stone towards fully resistant deep learning models. Code and pre-trained models are available at this https URL and this https URL.

Authors

Ludwig Schmidt

7 papers

Dimitris Tsipras

4 papers

A. Ma̧dry

3 papers

References45 items

Second-Order Adversarial Attack and Certifiable Robustness

Towards the first adversarially robust neural network model on MNIST

Decision-Based Adversarial Attacks: Reliable Attacks Against Black-Box Machine Learning Models

Ground-Truth Adversarial Examples

Analysis of classifiers’ robustness to adversarial perturbations

Towards Deep Learning Models Resistant to Adversarial Attacks

Published in

International Conference on Learning Representa...(2017)

External Links:

Generate Graph

TL;DR

Abstract

Authors

Ludwig Schmidt

7 papers

Dimitris Tsipras

4 papers

A. Ma̧dry

3 papers

References45 items

Second-Order Adversarial Attack and Certifiable Robustness

Towards the first adversarially robust neural network model on MNIST

Decision-Based Adversarial Attacks: Reliable Attacks Against Black-Box Machine Learning Models

Ground-Truth Adversarial Examples

Analysis of classifiers’ robustness to adversarial perturbations

Aleksandar Makelov

1 papers

Adrian Vladu

1 papers

Adversarial Example Defenses: Ensembles of Weak Defenses are not Strong

Adversarial Examples Are Not Easily Detected: Bypassing Ten Detection Methods

Ensemble Adversarial Training: Attacks and Defenses

The Space of Transferable Adversarial Examples

Feature Squeezing: Detecting Adversarial Examples in Deep Neural Networks

Towards Robust Deep Neural Networks with BANG

Robust Large Margin Approaches for Machine Learning in Adversarial Settings

Adversarial Machine Learning at Scale

Accessorize to a Crime: Real and Stealthy Attacks on State-of-the-Art Face Recognition

Towards Evaluating the Robustness of Neural Networks

On the Effectiveness of Defensive Distillation

Defensive Distillation is Not Robust to Adversarial Examples

Robust Large Margin Deep Neural Networks

Transferability in Machine Learning: from Phenomena to Black-Box Attacks using Adversarial Samples

Deep Residual Learning for Image Recognition

On Robust Optimization

The Limitations of Deep Learning in Adversarial Settings

DeepFool: A Simple and Accurate Method to Fool Deep Neural Networks

Distillation as a Defense to Adversarial Perturbations Against Deep Neural Networks

A Unified Gradient Regularization Family for Adversarial Examples

Learning with a Strong Adversary

Analysis of classifiers’ robustness to adversarial perturbations

Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification

Explaining and Harnessing Adversarial Examples

Towards Deep Neural Network Architectures Robust to Adversarial Examples

Deep neural networks are easily fooled: High confidence predictions for unrecognizable images

Intriguing properties of neural networks

Evasion Attacks against Machine Learning at Test Time

ImageNet classification with deep convolutional neural networks

A unified architecture for natural language processing: deep neural networks with multitask learning

Nightmare at test time: robust learning by feature deletion

Adversarial classification

Statistical Decision Functions

Statistical Decision Functions Which Minimize the Maximum Risk

Contributions to the Theory of Statistical Estimation and Testing Hypotheses

Adversarial Example Defense: Ensembles of Weak Defenses are not Strong

A lot of recent literature on adversarial training discusses the phenomenon of transferability Goodfellow et al

Understanding Adversarial Training: Increasing Local Stability of Neural Nets through Robust Optimization

Building on the above insights

Tensor ﬂow models repository

Field of Study

Computer ScienceMathematics

Journal Information

Name

ArXiv

Volume

abs/1706.06083

Venue Information

Name

International Conference on Learning Representations

Type

conference

URL

https://iclr.cc/

Alternate Names

Int Conf Learn Represent
ICLR