Neural Dynamics, Network States and Criticality

a) Information processing in networks close to a critical state

Being at the border between regular and chaotic activity, critical network states display interesting dynamical features such as power-law distributions in size and duration of synchronous events, and they provide a rich repertoire of collective states which are potentially useful for information processing. But what role does critical behavior really play if the brain actively processes stimulus information? 

We investigate this question by relating aspects of information processing in the visual system to the dynamics in neural networks close to a critical state. In particular, we focus on mechanisms that link local features in visual scenes into global figures such as contours or textures, and evaluate conditions under which a near-critical dynamics optimizes computational performance. We also study how top-down processes modulate information integration: By integrating physiological evidence with computational work, we have been able to show that selective attention might tune the visual system to a critical state and thereby enhance stimulus representation.

b) Transient dynamics and attentional modulation in visual area MT

The world is a dynamic place, and temporal changes in a scene contain important information relevant for behavior. On the neuronal level, such changes are often represented by transient increases or decreases in activity before firing rates settle into a sustained level. Our studies aim to uncover the mechanisms behind neural transients, and to quantify their impact on information processing. Hereby we closely collaborate with Detlef Wegener (Brain Research Institute, University of Bremen) and Jannis Hildebrandt (Neuroscience Department, University of Oldenburg), to tightly link computational work with physiological data. Currently, we investigate a neural circuit with DIVisive INhibition on Excitatory units (DIVINE-model - panel A), which nicely captures single-unit and population responses of neurons in area MT (panel B and C) to drifting visual stimuli subjected to abrupt changes in drift velocity. When attentional modulation is added as a multiplicative input change to our model, it turns out that the dynamics of the transients, but not the sustained states, correlate nicely with behavior.