On the high rate, independence, and optimal prediction assumptions in predictive coding

Abstract

We examine the common assumptions of high rate, independence, and optimal prediction often used in the theoretical performance analyses of differential predictive coding. We demonstrate that these assumptions are never valid for this structure. We then perform an analysis of the performance of differential predictive coding without these assumptions and show that not only are these assumptions invalid, the results obtained using these assumptions underestimate system performance. Both theoretical and practical illustrations are provided.