Abstract
If maps of the sky observed at different wavelengths are available , it is possible to combine them to reveal the " cosmological microwave background ": a snapshot of the primordial Universe. This operation is an example of " component separation ": the art and practice of extracting underlying elementary signals from multiple observations.
The seminar focuses on Independent Component Analysis (ICA): a separation method based solely on the assumption of independent, non-Gaussian underlying components. The likelihood analysis of ICA reveals some key notions of information theory: entropy, Kullback divergence, mutual information... We'll see how these quantities are naturally linked in a geometric interpretation: that of information geometry, built on Kullbackdivergence and whose metric is Fisher information.
