Random Matrix
Random matrix theory (RMT) provides a powerful mathematical framework for analyzing the behavior of large, high-dimensional matrices, particularly those arising in machine learning and data analysis. Current research focuses on applying RMT to understand the performance of various algorithms, including kernel ridge regression, spectral clustering, and deep learning models, often examining the spectral properties of weight matrices and kernels in these contexts. This work yields insights into phenomena like double descent, generalization error, and the impact of model architecture and hyperparameters, ultimately improving algorithm design and predictive power in high-dimensional settings.
Papers
A Random Matrix Approach to Low-Multilinear-Rank Tensor Approximation
Hugo Lebeau, Florent Chatelain, Romain Couillet
Deep Equilibrium Models are Almost Equivalent to Not-so-deep Explicit Models for High-dimensional Gaussian Mixtures
Zenan Ling, Longbo Li, Zhanbo Feng, Yixuan Zhang, Feng Zhou, Robert C. Qiu, Zhenyu Liao