Universal lossy compression under logarithmic loss
Yanina Y. Shkel, Maxim Raginsky, Sergio Verdú · 2017
Universal lossy source coding with the logarithmic loss distortion criterion is studied. Bounds on the non-asymptotic fundamental limit of fixed-length universal coding with respect to a family of distributions are derived. These bounds generalize the well-known minimax bounds for universal lossless source coding. The asymptotic behavior of the resulting optimization problem is studied for a family of i.i.d. sources with a finite alphabet size, and is characterized up to a constant. The redundancy of memoryless sources behaves like k/2 log n, where n is the blocklength and k is the number of degrees of freedom in the parameter space. The impact of the coding rate is on the constant term: higher compression rate effectively reduces the volume of the parameter uncertainty set.