Phase transitions of the information distortion
Albert Parker, Department of Mathematics, Montana State University
We analyze the phase transitions of a class of annealing
problems which originate from Rate Distortion Theory
and the Deterministic Annealing approach to clustering.
Solutions of these problems have been used to
approximate neural coding schemes in simple sensory systems.
The phase transitions or bifurcations that the solutions undergo in the
annealing process can be analyzed by
capitalizing on the symmetries of the solutions. Theoretical and
numerical results are presented for the Information Distortion problem.