Spectral-Transport Stability and Benign Overfitting in Interpolating Learning

Collection of new machine-learning papers (arXiv stat/ML) covering theoretical and methodological advances: a spectral-transport stability framework characterizing benign vs. destructive overfitting in interpolating estimators; a two-stage, tensor-completion-based design for large-scale factorial experiments; H-MRS for causal DAG learning on positive-valued data; an online projected functional-gradient algorithm (P-FGD) for nonparametric quantile regression; SPICE identifiability and SPICE-Net for causal effect estimation using a single proxy; a predictive-first, beam-style streaming HMMs alg­