Reproduce “Increasing Missingness to Reduce Bias: Richardson-SGD with Missing Data” 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
Ferdinand Genans, Erwan Scornet
The authors introduce Richardson-SGD to reduce gradient bias when training covariates are missing. It adds controlled thinning to observed entries, computes gradients at two missingness levels using one shared imputation, and combines them to cancel leading bias. Under independent hMCAR or scalable MAR masks, expected bias falls from first to second order; this does not cover arbitrary dependent masks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.