Reproduce “One-Shot Learning in Discriminative Neural Networks” 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
Jordan Burgess, James Robert Lloyd, Zoubin Ghahramani
The authors adapt a pretrained convolutional network to classify new categories from few examples. They keep image features fixed and use a Gaussian prior from existing classifier weights to guide new-class weights. On five-way, one-shot CIFAR-100 classification, it reached almost 40% accuracy, versus 34% for cosine nearest neighbors and 30% for trained softmax, with evaluation limited to 20 examples per class.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.