Reproduce “Adversarial Attacks are a Surprisingly Strong Baseline for Poisoning Few-Shot Meta-Learners” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Elre T. Oldewage, John Bronskill, Richard E. Turner
The authors introduce adversarial support poisoning (ASP) for few-shot image classification, jointly perturbing labeled support images through task adaptation to reduce accuracy on seed and unseen query sets. On CIFAR-100, ASP reduces accuracy from 53.5% clean to 7.7% on unseen queries, but transfer between feature extractors does not significantly outperform the cheaper swap attack in reported comparisons.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.