Reproduce “B2P-Corr: Batch-to-Population Gradient Estimators for Non-Decomposable Correlation Losses” 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
Jingquan Yan, Yuwei Miao, Peiran Yu, Junzhou Huang
B2P-Corr trains regression models for correlation and concordance objectives using small batches. It maintains estimates of moments and uses them as coefficients in pseudo-losses; its variants use exponential moving averages or pairs with reservoir stabilization. On KoNViD-1k at batch size 2, B2P-EMA raises PCC from 0.865 to 0.895; this result does not establish moment bias as the only bottleneck.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.