arXiv subjects

Nicole Finnie

Publications and source records attributed to Nicole Finnie.

3 recordsLinked to original sources

arXiv2022

Give Me Your Attention: Dot-Product Attention Considered Harmful for Adversarial Patch Robustness

Neural architectures based on attention such as vision transformers are revolutionizing image recognition. Their main benefit is that attention allows reasoning about all parts of a scene jointly. In this paper, we show how the global reasoning of (scaled) dot-product attention can be the source of a major vulnerability when confronted with adversarial patch attacks. We provide a theoretical understanding of this vulnerability and relate it to an adversary's ability to misdirect the attention of all queries to a single key token under the control of the adversarial patch. We propose novel adversarial objectives for crafting adversarial patches which target this vulnerability explicitly. We show the effectiveness of the proposed patch attacks on popular image classification (ViTs and DeiTs) and object detection models (DETR). We find that adversarial patches occupying 0.5% of the input can lead to robust accuracies as low as 0% for ViT on ImageNet, and reduce the mAP of DETR on MS COCO to less than 3%.

cs.CV↗

arXiv2021

Meta Adversarial Training against Universal Patches

Recently demonstrated physical-world adversarial attacks have exposed vulnerabilities in perception systems that pose severe risks for safety-critical applications such as autonomous driving. These attacks place adversarial artifacts in the physical world that indirectly cause the addition of a universal patch to inputs of a model that can fool it in a variety of contexts. Adversarial training is the most effective defense against image-dependent adversarial attacks. However, tailoring adversarial training to universal patches is computationally expensive since the optimal universal patch depends on the model weights which change during training. We propose meta adversarial training (MAT), a novel combination of adversarial training with meta-learning, which overcomes this challenge by meta-learning universal patches along with model training. MAT requires little extra computation while continuously adapting a large set of patches to the current model. MAT considerably increases robustness against universal patch attacks on image classification and traffic-light detection.

cs.LG↗

arXiv2021

The Tracking Machine Learning challenge : Accuracy phase

This paper reports the results of an experiment in high energy physics: using the power of the "crowd" to solve difficult experimental problems linked to tracking accurately the trajectory of particles in the Large Hadron Collider (LHC). This experiment took the form of a machine learning challenge organized in 2018: the Tracking Machine Learning Challenge (TrackML). Its results were discussed at the competition session at the Neural Information Processing Systems conference (NeurIPS 2018). Given 100.000 points, the participants had to connect them into about 10.000 arcs of circles, following the trajectory of particles issued from very high energy proton collisions. The competition was difficult with a dozen front-runners well ahead of a pack. The single competition score is shown to be accurate and effective in selecting the best algorithms from the domain point of view. The competition has exposed a diversity of approaches, with various roles for Machine Learning, a number of which are discussed in the document

hep-ex↗