The Regularization Parameter: Sparse Precision Matrix Estimation

This document is an arXiv stat_ml feed (multiple 2026-07-10 postings) collecting new papers on high-dimensional statistics, probabilistic modelling, and reinforcement learning. Highlights include: a closed-form, matrix-valued regularization parameter for ℓ1-regularized Gaussian precision estimation that replaces cross-validation and yields consistency and sparsistency; a positive semi-definite kernel-density approach (PSD Impute) that frames MCAR imputation as convex density estimation with closed-form marginals; analysis of diffusion policies identifying a drift Lipschitz budget K that trades