Reproduce “Learning from Label Proportions: A Mutual Contamination Framework” 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
Clayton Scott, Jianxin Zhang
The authors introduce LMMCM for learning individual labels from unlabeled bags with known proportions. It treats bags with different proportions as contaminated views of shared positive and negative classes, correcting the loss to estimate balanced error rate (BER), the equally weighted average of class-conditional losses. On Adult and MAGIC data, AUC stayed higher at larger bags, without isolating component effects.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.