Reproduce “AdaFlood: Adaptive Flood Regularization” 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

Wonho Bae, Yi Ren, Mohamad Osama Ahmed, Frederick Tung, Danica J. Sutherland, Gabriel L. Oliveira

AdaFlood assigns each training example its own flooding target, estimated from held-out auxiliary-model predictions, rather than using one shared target. On CIFAR-100 image classification with ResNet18 and L₂ regularization, it reports 62.31% ± 0.14 accuracy, compared with 61.63% ± 0.12 for iFlood, across three runs. The authors report no significant advantage over the unregularized baseline on SVHN.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.