Reproduce “Training Confidence-calibrated Classifiers for Detecting Out-of-Distribution Samples” 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
Kimin Lee, Honglak Lee, Kibok Lee, Jinwoo Shin
The authors train classifiers to be less confident on generated examples near the boundary of familiar image data, while retaining supervised classification. In VGGNet-13, with CIFAR-10 in-distribution and LSUN out-of-distribution, joint confidence loss raises true negative rate at 95% true positive rate from 14.0% to 39.1%. These results cover tested image datasets, not arbitrary unknown distributions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.