Reproduce “Uncertainty Estimation Using a Single Deep Deterministic Neural Network” 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
Joost van Amersfoort, Lewis Smith, Yee Whye Teh, Yarin Gal
DUQ estimates whether an input fits any learned class using one model and one test-time forward pass. It compares input features with class prototypes and uses distance from the nearest prototype as an uncertainty signal, with a two-sided input-gradient penalty. On FashionMNIST versus MNIST, it reaches AUROC 0.955, above a five-model ensemble’s 0.861 but below a generative method’s 0.994.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.