A new class of learning algorithms for independent component analysis (ICA) is presented. Starting from theoretical discussions on convex divergence, this information measure is minimized to derive new ICA algorithms. Since the convex divergence includes logarithmic information measures as special cases, the presented method comprises faster algorithms than existing logarithmic ones. Another important feature of this paper's ICA algorithm is to accept supervisory information. This ability is utilized to reduce the permutation indeterminacy which is inherent in usual ICA. By this method, the most important activation pattern can be found as the top one. The total algorithm is tested through applications to brain map distillation from functional MRI data. The derived algorithm is faster than logarithmic ones with little additional memory requirement, and can find task related brain maps successfully via conventional personal computer.
|Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
|Published - 2003
ASJC Scopus subject areas