1: \begin{abstract}

2:     The principal component analysis (PCA) is a powerful standard tool for

3: 		reducing the dimensionality of data.

4: 		Unfortunately, it is sensitive to outliers so that various robust PCA variants 

5: 		were proposed in the literature.

6: 		This paper addresses the robust PCA  

7: 		by successively determining the directions of lines

8: 		having minimal Euclidean distances

9: 		from the data points.

10: 		The corresponding energy functional is not differentiable at a finite number of directions

11: 		which we call anchor directions. 

12: 		We derive a Weiszfeld-like algorithm for minimizing the energy functional

13: 		which has several advantages over existing algorithms.

14: 		Special attention is paid to the careful handling of the anchor directions, where

15: 		we take the relation between local minima and 

16: 		one-sided derivatives of Lipschitz continuous functions on submanifolds of $\mathbb R^d$ 

17: 		into account.

18: 		Using ideas for stabilizing the classical Weiszfeld algorithm at anchor points and the

19: 		Kurdyka–Łojasiewicz property of the energy functional,

20: 		we prove global convergence of the whole sequence of iterates generated by the algorithm to a

21: 		critical point of the energy functional.

22: 		Numerical examples demonstrate the very good performance of our algorithm.

23: \end{abstract}

24: