Since 01.04.2011 the Max Planck Emeritus group of Wolf Singer is affiliated also with the Ernst Strüngmann Institute that currently develops at the site occupied previously by the Max Planck Institute for Brain Research on the campus of the Medical Faculty.
Research focuses on the analysis of neuronal processes in the mammalian cerebral cortex that underlie higher cognitive functions. We pursue the hypothesis that the evolution of neocortex introduced novel strategies of information processing that go beyond the classical strategies of rate and labelled line coding in hierarchical processing architectures. We consider cortex as a delayed coupled recurrent network whose nodes consist of feature selective oscillators. Information about the statistical contingencies among features of perceptual objects (the priors) is supposed to be stored in the weight distributions of the recurrent connections (edges) between the nodes. Such recurrent networks exhibit exceedingly complex non-linear dynamics that can in principle be used as basis for computations in high dimensional state space. The aim of our investigations is to examine whether nature exploits this option. One of the core predictions is that relevant information is encoded in the precise temporal relations (correlation structure) between the discharges of distributed neurons. We assume that network oscillations, synchronisation, phase shifts and cross frequency coupling provide the temporal structure of neuronal responses required for the dynamic encoding of relations. In order to examine time varying correlation structures in the activity of distributed neurons we obtain massive parallel recordings from multiple cortical sites and analyse the resulting high dimensional response patterns with advanced methods of time series analysis and machine learning techniques. In most of the projects we use the visual system of behaviourally trained non-human primates to establish relations between neuronal response vectors, specific stimulation conditions and selected cognitive functions. These functions comprise perceptual decisions, control of attention, short term memory and predictive coding. Because the analysis of the high dimensional, non stationary response vectors requires advanced computational methods from the fields of non-linear dynamics, complexity theory and machine learning, we cooperate and share our data with colleagues from these disciplines working at the Frankfurt Institute for Advanced Studies (FIAS) and in several European countries.
For summaries of the conceptual background motivating our research see
- Singer: TICS 17: 616-626 (2013) Link
- Singer & Lazar: Front. Comput. Neurosci. 10: 99 (2016) Link