Reproduce “The Pursuit of Human Labeling: A New Perspective on Unsupervised Learning” 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
Artyom Gadetsky, Maria Brbic
HUME searches for image groupings that generalize across two pretrained feature spaces. It learns prototypes in one frozen representation and updates candidate labels using held-out performance of a linear classifier in the other. On CIFAR-10, it reports 88.4% ACC and 77.6% ARI, versus SPICE’s 84.5% and 70.9%. It also outperforms compared unsupervised baselines on STL-10, CIFAR-100-20, and ImageNet-1000.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.