High-Dimensional Partial Least Squares: Spectral Analysis and Fundamental Limitations
The episode discusses Leger and Chatelain's paper on High-Dimensional Partial Least Squares using random matrix theory to analyze its spectral properties and limitations. Hosts discuss how this framework provides a mathematical reason for PLS performance, suggesting methods to set thresholds for reliable signal detection, quantify alignment between estimated and true directions, and develop noise-aware filtering mechanisms.