Rate distortion theory with generalized information measures via convex programming duality

A new generalized average mutual information measure (GAMIM) is introduced in terms of Csiszar \phi -divergence, and the associated rate distortion function R_{\phi} is studied. The main objective is to derive in a unified way a dual representation of R_{\phi} , then to use it to generalize classical results (corresponding to \phi(t) = t \log t ) in rate distortion theory and extend results associated with other concepts of GAMIM. Our development uses the methodology of convex programming duality extensively.