Poisoning Web-Scale Training Datasets is Practical

arXiv:2302.10149v2 Announce Type: replace-cross
Abstract: Deep learning models are often trained on distributed, web-scale datasets crawled from the internet. In this paper, we introduce two new dataset poisoning attacks that intentionally introduce malicious examples to a model's performance. Our attacks are immediately practical and could, today, poison 10 popular datasets. Our first attack, split-view poisoning, exploits the mutable nature of internet content to ensure a dataset annotator's initial view of the dataset differs from the view downloaded by subsequent clients. …

abstract arxiv attacks cs.cr cs.lg dataset datasets deep learning distributed examples internet paper performance poisoning attacks popular practical scale s performance split training training datasets type view web