A1124
Title: Asymptotics in high dimensional regression under dependence
Authors: Samriddha Lahiry - National University of Singapore (Singapore) [presenting]
Abstract: Over the past decade, a novel regime has emerged in supervised learning, wherein both the number of features p and the number of samples n diverge with their ratio converging to a positive constant. The widespread adoption of this paradigm, also known as the proportional asymptotic regime in statistical research, can, in part, be attributed to its precise asymptotic results, which demonstrate strong performance even on finite samples. Moreover, this framework relaxes the need for stringent sparsity assumptions on the underlying signals, allowing for structured classes of both dense and sparse signals. However, even in high-dimensional regression, existing techniques cannot be readily extended to accommodate dependence structures, thereby posing non-trivial challenges. Two models of dependence are investigated, and precise asymptotic results are derived for each.
