Collaborating Authors

A Simple Sticker Tricked Neural Networks Into Classifying Anything as a Toaster


Image recognition technology may be sophisticated, but it is also easily duped. Researchers have fooled algorithms into confusing two skiers for a dog, a baseball for espresso, and a turtle for a rifle. But a new method of deceiving the machines is simple and far-reaching, involving just a humble sticker.

On Physical Adversarial Patches for Object Detection Machine Learning

In this paper, we demonstrate a physical adversarial patch attack against object detectors, notably the YOLOv3 detector. Unlike previous work on physical object detection attacks, which required the patch to overlap with the objects being misclassified or avoiding detection, we show that a properly designed patch can suppress virtually all the detected objects in the image. That is, we can place the patch anywhere in the image, causing all existing objects in the image to be missed entirely by the detector, even those far away from the patch itself. This in turn opens up new lines of physical attacks against object detection systems, which require no modification of the objects in a scene. A demo of the system can be found at

How to Hack an Intelligent Machine


This week Microsoft and Alibaba stoked new fears that robots will soon take our jobs. The two companies independently revealed that their artificial intelligence systems beat humans at a test of reading comprehension. The test, known as the Stanford Question Answering Dataset (SQuAD), was designed to train AI to answer questions about a set of Wikipedia articles.

Attribution-driven Causal Analysis for Detection of Adversarial Examples Machine Learning

Attribution methods have been developed to explain the decision of a machine learning model on a given input. We use the Integrated Gradient method for finding attributions to define the causal neighborhood of an input by incrementally masking high attribution features. We study the robustness of machine learning models on benign and adversarial inputs in this neighborhood. Our study indicates that benign inputs are robust to the masking of high attribution features but adversarial inputs generated by the state-of-the-art adversarial attack methods such as DeepFool, FGSM, CW and PGD, are not robust to such masking. Further, our study demonstrates that this concentration of high-attribution features responsible for the incorrect decision is more pronounced in physically realizable adversarial examples. This difference in attribution of benign and adversarial inputs can be used to detect adversarial examples. Such a defense approach is independent of training data and attack method, and we demonstrate its effectiveness on digital and physically realizable perturbations.

Breaking neural networks with adversarial attacks


As many of you may know, Deep Neural Networks are highly expressive machine learning networks that have been around for many decades. In 2012, with gains in computing power and improved tooling, a family of these machine learning models called ConvNets started achieving state of the art performance on visual recognition tasks. Up to this point, machine learning algorithms simply didn't work well enough for anyone to be surprised when it failed to do the right thing. In 2014, a group of researchers at Google and NYU found that it was far too easy to fool ConvNets with an imperceivable, but carefully constructed nudge in the input. Let's look at an example.