Robust Multi-Task Learning for Principal Component Analysis
New multi-task PCA procedures are proposed that exploit similarity structure across tasks to improve eigenspace estimation while remaining robust to outlier tasks, addressing a key challenge in robust multi-task learning.