Reproduce “From internal representations to model improvement through prediction errors” 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
Yushi Nakaya, Kenichi Higuchi, Shuichi Ishida
The authors present a method that selects unlabeled object-detection images by using the detector’s features and predictions to estimate likely error types and their effects on average precision. At 15% labeled data, it achieved higher mAP than rarity-only selection on BDD100K and COCO 2017, although rarity-only selection found more true errors. Across four conditions, it beat rarity-only selection but did not consistently beat other methods.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.