An improved stochastic gradient algorithm for principal component analysis and subspace tracking
Jeroen Dehaene, Marc Moonen, Joos P. L. Vandewalle · IEEE Transactions on Signal Processing · 1997
We propose a new stochastic gradient algorithm for principal component analysis and subspace tracking, requiring O(nm) operations per update, where n is the number of input signals, and m is the signal subspace dimension. A parallel version with problem size independent throughput is obtained at the expense of O(n/sup 2/) additional flops.