1: \begin{abstract}

2: \trarxiv{\noindent\sloppy \textbf{Abstract}\\[0.5cm]}{}

3: %

4: \noindent%

5: Fully symmetric learning rules for principal component analysis can be

6: derived from a novel objective function suggested in our previous

7: work. We observed that these learning rules suffer from slow

8: convergence for covariance matrices where some principal eigenvalues

9: are close to each other. Here we describe a modified objective

10: function with an additional term which mitigates this convergence

11: problem. We show that the learning rule derived from the modified

12: objective function inherits all fixed points from the original

13: learning rule (but may introduce additional ones). Also the stability

14: of the inherited fixed points remains unchanged. Only the steepness of

15: the objective function is increased in some directions. Simulations

16: confirm that the convergence speed can be noticeably improved,

17: depending on the weight factor of the additional term.

18: 

19: \trarxiv{%

20:   \vspace*{1cm}

21:   Please cite as: Ralf Möller. {\em \trtitle}. Technical Report, Computer

22:   Engineering, Faculty of Technology, Bielefeld University, \tryear,

23:   version of \today, \url{www.ti.uni-bielefeld.de}.

24: }{%

25: }

26: \end{abstract}

27: