Towards Deep Learning Models Resistant to Adversarial Attacks
By Aleksander Madry, Aleksandar Makelov, Ludwig Schmidt, Dimitris Tsipras, Adrian Vladu. 2017
15,165 citations recorded by the source corpus.
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.
Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Safety, Robustness & Interpretability
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
- Yoshua Bengio
- Percy Liang
- Mohit Bansal
- Quoc V. Le
- Jianfeng Gao
- Sergey Levine
- Ruslan Salakhutdinov
- Luke S. Zettlemoyer
In Computer Vision and Pattern Recognition
- Luc Van Gool
- Dacheng Tao
- Trevor Darrell
- Ming-Hsuan Yang
- Xiaogang Wang
- Bernt Schiele
- Jitendra Malik
- Alan L. Yuille
In Safety, Robustness & Interpretability
- Matthias Hein
- Nicholas Carlini
- Nicolas Papernot
- Aleksander Madry
- Florian Tramèr
- Huan Zhang
- Cho-Jui Hsieh
- Bo Li
Citations in the corpus
- Adversarial Training for Large Neural Language Models (2020, methodology): “Adversarial training (Madry et al., 2017; Good-fellow et al., 2014) has been well studied in computer vision, but past work shows that it often hurts generalization (Raghunathan et al., 2019; Min et al., 2020).”
- Shortcut learning in deep neural networks (2020, background): “ent in most data and rarely disappear by adding more data [32,69,56,38,33]. Modifying the training data to block specific shortcuts has been demonstrated to work for reducing adversarial vulnerability [110] and texture bias [38]. Goal: loss function. The most commonly used loss function, cross-entro”
- Exploring Self-Attention for Image Recognition (2020, methodology): “We subject the trained models to white-box targeted PGD attacks [24].”
- Self-Training With Noisy Student Improves ImageNet Classification (2019, neutral): “iterations [43], which is far from the SOTA results.”
- Measuring Robustness to Natural Distribution Shifts in Image Classification (2020, methodology): “Encouragingly, the past few years have seen substantial progress in robustness to these distribution shifts, e.g., see [13, 28, 34, 55, 57, 66, 93, 96, 105, 114, 115] among many others.”
- The Many Faces of Robustness: A Critical Analysis of Out-of-Distribution Generalization (2020, background): “The space of augmentations that help robustness includes various types of noise [22, 27, 21], highly unnatural image transformations [10, 38, 39], or compositions of simple image transformations such as Python Imaging Library operations [4, 15].”
- Large-Scale Adversarial Training for Vision-and-Language Representation Learning (2020, background): “For image modality, most previous work observes that robustness is at odds with generalization, i.e. , trained models are able to resist adversarial attacks on clean images at the expense of performance [42, 72, 83].”
- Natural Adversarial Examples (2019, methodology): “While we aim to estimate the worst-case accuracy in natural settings, most work studies `p adversarial attacks [36].”
- Benchmarking Neural Network Robustness to Common Corruptions and Perturbations (2019, background): “…new defenses (Lu et al., 2017; Papernot et al., 2017; Metzen et al., 2017; Hendrycks & Gimpel, 2017a) often quickly succumb to new attacks (Evtimov et al., 2017; Carlini & Wagner, 2017; 2016), with some exceptions for perturbations on small images (Schott et al., 2018; Madry et al., 2018).”
- Do ImageNet Classifiers Generalize to ImageNet? (2019, methodology): “[41] Aleksander Madry, Aleksandar Makelov, Ludwig Schmidt, Dimitris Tsipras, and Adrian Vladu.”
- Towards Robust Vision Transformer (2021, methodology): “For evaluating the adversarial robustness, we adopt single-step attack algorithm FGSM [11] and multi-step attack algorithm PGD [26] with steps t = 5 , step size α = 0 .”
- CutMix: Regularization Strategy to Train Strong Classifiers With Localizable Features (2019, background): “One straightforward way to enhance robustness and uncertainty is an input augmentation by generating unseen samples [25].”
- Your Classifier is Secretly an Energy Based Model and You Should Treat it Like One (2019, background): “With respect to the L∞ norm, JEM performs competitive with, but slightly below state-of-the-art adversarial training (Madry et al., 2017).”
- Scalable agent alignment via reward modeling: a research direction (2018, background): “…the effect of adversarially crafted inputs to neural networks (Szegedy et al., 2013), so far the empirically most effective strategy has been adversarial training : training the model explicitly on adversarially perturbed inputs (Madry et al., 2017; Uesato et al., 2018; Athalye et al., 2018).”
- Repurposing Pretrained Models for Robust Out-of-domain Few-Shot Learning (2021, background): “While stronger attacks have been proposed (such as using projected gradient descent (Madry et al., 2018)), we will focus on the FGSM approach in this paper for simplicity.”
- Online Adversarial Attacks (2021, methodology): “As representative attack strategies, we use the well-known FGSM attack [Goodfellow et al., 2015] and a universal whitebox attack in PGD [Madry et al., 2017].”
- Adversarial Example Games (2020, methodology): “Under standard PGD adversarial training [51] this corresponds to a worst-case setting for our attacker.”