Authorization to Lend and Reproduce the Thesis

This work is concerned with the applicability of mutual information as a measure of relevance in neural coding. We give a detailed overview of common approaches to the application of information-theoretic measures in neuroscience. In order to create an environment that allows for the analysis of mutual information, we generate synthetic spike trains based on stimulus models that are easy to describe but sufficiently powerful to be relevant in practice. Different methods to extract a random variable suitable for the estimation of mutual information from spike trains are discussed and their shortcomings illuminated. We go on and demonstrate the sensitivity of mutual information to an inappropriate choice of parameters in the extraction of the random variable and show how the temporal characteristics of the stimulus have to be factored in. Finally, we illustrate the detrimental effects on certain estimators of mutual information that occur if insufficient amounts of data are available.