Reproduce “Data-Efficient Classifier Pretraining Through Diffusion Models” 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
Qinyu Li, Yee Whye Teh, Chris Holmes, Joel Jaskari
DiffPT pretrains a classifier backbone by teaching it to predict Gaussian noise added to inputs at varying levels, then fine-tunes that backbone for classification using the same dataset, without unlabeled data. On EyePACS classification with 1% of the training split, DiffPT reaches 0.744 AUROC, versus 0.643 for SimCLR and 0.618 from scratch; this result does not establish superiority across tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.