Analysis of neural signals: Interdependence, information coding, and relation to network models

Most of the functionality of the brain is thought to emerge fr om communications between neurons using spikes. Thus, for investigating neur al f nction it is essential to monitor spiking activity of neurons. Experimentally mea sured signals, however, do often not directly reflect spiking activity of neurons, but i nstead comprise a mixture of biophysical events from various origins. In this dissert ation I investigate the interdependence of two signals: extracellularly measured spiki ng activity and local field potentials. First, by means of machine learning techniques I ask to what d egree spike trains can be inferred from simultaneously measured local field potent ials from primary visual cortex and lateral geniculate nucleus of non-anesthetized and anesthetized macaque monkeys. Second, using an information theoretic approach I further s ow that in primary visual cortex of macaque monkeys, spikes are related to LFP osc illations in a stimulus dependent manner. In particular, information about a scene in a movie can be predicted with higher precision from spike trains, when the pha se of local field potential oscillations is taken into account. The structure of experimental spike trains in response to na tural scenes is rich, with periods of reliable high activity bursts intermingled with long silent periods. I develop a neural network model based on many anatomical particulari ties of the primary visual cortex of macaques in order to compare the statistics of the s pik trains generated by an artificial neural network under similar stimulus condition s. To achieve a close match to the data free parameters of the model are optimized using a ne w method for comparing multi-dimensional distributions, called Maximum Mean Dis crepancy ( MMD).