An expectation-maximization approach to nonlinear component analysis

The proposal of considering nonlinear principal component analysis as a ker nel eigenvalue problem has provided an extremely powerful method of extract ing nonlinear features for a number of classification and regression applic ations. Whereas the utilization of Mercer kernels makes the problem of comp uting principal components in, possibly, infinite-demensional feature space s tractable, there are still the attendant numerical problems of diagonaliz ing large matrices. In this contribution, we propose an expectation-maximiz ation approach for performing kernel principal component analysis and show this to be a computationally efficient method, especially when the number o f data points is large.