B1796
Title: Optimal penalty selection for high-dimensional covariance matrices with an application in NLP
Authors: El Mehdi Issouani - University Paris Nanterre (France) [presenting]
Patrice Bertail - University of Paris-Ouest-Nanterre-La Defense (France)
Emmanuelle Gautherat - University of Reims (France)
Abstract: The Hotelling $T^2$ statistics is considered in a large-dimension framework, replacing the covariance matrix with a penalized version. In the same spirit of the approach taken in a past study for regularizing the covariance matrix, an optimal penalty coefficient selection is proposed. This enables the establishment of controls for Penalized Hotelling $T^2$ statistics in high-dimensional settings. During this presentation, the significance of this penalty method is highlighted and a geometric interpretation is provided involving the projection of the problem into a Hilbert space, concluding with a case study in natural language processing.