Information-theoretic analysis of signal processing systems: application to neural coding

We analyze the signal processing capabilities of a system given access to its inputs and outputs, but not assuming any special structure other than they are stochastic. The analysis techniques are based on information-theoretic distance measures and on empirical theories derived from work in universal signal processing. We apply our techniques to the analysis of single- and multi-neuron discharge patterns, finding that neurons can encode multiple attributes simultaneously and in a time-varying fashion