Characterization of Gaussian Universality Breakdown in High-Dimensional Empirical Risk Minimization
Researchers have developed a new theoretical framework to understand how statistical learning models behave with non-Gaussian data. The study extends the Convex Gaussian Min-Max Theorem to analyze high-dimensional empirical risk minimization, providing an asymptotic characterization of key statistics like the estimator's mean and covariance. This work clarifies the boundaries of Gaussian universality in empirical risk minimization and offers insights into the asymptotic equivalence of regularizers. AI