Missing-Data-Induced Phase Transitions in Spectral Partial Least Squares
This study investigates how missing data in two different views of Partial Least Squares (PLS) spectral analysis affects its performance, specifically focusing on a high-dimensional spiked model. It found that dual missingness reduces the effective signal strength by a factor related to the joint retention probability, shifting the critical threshold for successful recovery and providing exact mathematical formulas for overlap predictions.