APPLICATION OF ENSEMBLE LEARNING ICA TO INFRA-RED IMAGING
James W. Miskin and David J. C. MacKay
jwm1003@mrao.cam.ac.uk
We apply Independent Component Analysis methods
to reduce a large data set of 2160 spectra. The data
were obtained by scanning an Infra-Red (THz) laser
beam across a cross section of a tooth. We show that a
Bilinear Decomposition can represent the 2160 spectra
by a reduced basis of around 10 latent spectra. We
also show that by selectively enforcing positivity of the
variables in the Bilinear Decomposition we obtain more
meaningful images. The method allows identi cation
of di erent regions of the tooth.