Timezone: »

MetaPoison: Practical General-purpose Clean-label Data Poisoning
W. Ronny Huang · Jonas Geiping · Liam Fowl · Gavin Taylor · Tom Goldstein

Wed Dec 09 09:00 AM -- 11:00 AM (PST) @ Poster Session 3 #1080

Data poisoning---the process by which an attacker takes control of a model by making imperceptible changes to a subset of the training data---is an emerging threat in the context of neural networks. Existing attacks for data poisoning neural networks have relied on hand-crafted heuristics, because solving the poisoning problem directly via bilevel optimization is generally thought of as intractable for deep models. We propose MetaPoison, a first-order method that approximates the bilevel problem via meta-learning and crafts poisons that fool neural networks. MetaPoison is effective: it outperforms previous clean-label poisoning methods by a large margin. MetaPoison is robust: poisoned data made for one model transfer to a variety of victim models with unknown training settings and architectures. MetaPoison is general-purpose, it works not only in fine-tuning scenarios, but also for end-to-end training from scratch, which till now hasn't been feasible for clean-label attacks with deep nets. MetaPoison can achieve arbitrary adversary goals---like using poisons of one class to make a target image don the label of another arbitrarily chosen class. Finally, MetaPoison works in the real-world. We demonstrate for the first time successful data poisoning of models trained on the black-box Google Cloud AutoML API.

Author Information

W. Ronny Huang (Google Research)
Jonas Geiping (University of Siegen)

Hello, I’m Jonas . I conduct research in computer science as PhD student at the University of Siegen. My background is in Mathematics, more specifically in mathematical optimization and I am interested in research that intersects current deep learning and mathematical optimization, with my main area of applications being computer vision.

Liam Fowl (University of Maryland)
Gavin Taylor (US Naval Academy)
Tom Goldstein (University of Maryland)

More from the Same Authors