Adversarial Vision Challenge

By Wieland Brendel, Jonas Rauber, Alexey Kurakin, Nicolas Papernot, Behar Veliqi, Marcel Salathé, Sharada Prasanna Mohanty, Matthias Bethge. 2018

61 citations recorded by the source corpus.

This competition was meant to facilitate measurable progress towards robust machine vision models and more generally applicable adversarial attacks. It encouraged researchers to develop query-efficient adversarial attacks that can successfully operate against a wide range of defenses while just observing the final model decision to generate adversarial examples. Conversely, the competition encouraged the development of new defenses that can resist a wide range of strong decision-based attacks. In this chapter we describe the organisation and structure of the challenge as well as the solutions developed by the top-ranking teams.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

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 Computer Vision and Pattern Recognition

Citations in the corpus

Excessive Invariance Causes Adversarial Vulnerability (2018, background): “Perturbation-based adversarial examples are appealing because they allow to quantitatively measure notions of adversarial robustness (Brendel et al., 2018).”
Bag of Tricks for Adversarial Training (2020, background): “(Brendel et al., 2020).”
Decoupling Direction and Norm for Efficient Gradient-Based L2 Adversarial Attacks and Defenses (2018, methodology): “(distance to a uniform gray image, as in [3]).”
Better Diffusion Models Further Improve Adversarial Training (2023, background): “, 2020) and dominated the winner solutions in adversarial competitions (Kurakin et al., 2018; Brendel et al., 2020).”
Black-Box Adversarial Attack with Transferable Model-based Embedding (2019, methodology): “In fact, in the recent NeurIPS 2018 Adversarial Vision Challenge (Brendel et al., 2018), many teams transferred adversarial examples from a source network as the starting point to carry out black-box boundary attack (Brendel et al.”
Boosting Adversarial Training with Hypersphere Embedding (2020, background): “Adversarial training (AT) is one of the most effective and widely studied defense strategies against adversarial vulnerability (Kurakin et al., 2018; Carlini et al., 2019; Brendel et al., 2020).”
Benchmarking Adversarial Robustness on Image Classification (2020, methodology): “The accuracy of a defense against an attack for a given perturbation budget [29] and the minimum distance of the adversarial perturbation [4] are used as the primary evaluation metrics, which are often insufficient to totally characterize the behaviour of the attacks and defenses.”
Curls & Whey: Boosting Black-Box Adversarial Attacks (2019, methodology): “As for Tiny-Imagenet with 200 image categories, we choose 2000 images, 10 images for each category.”