a) Selective attention and parallel functional configuration of neural systems
Selective attention serves to process behaviorally relevant parts of visual scenes while suppressing others. Our lab investigates selective information processing in the early visual system by combining numerical simulation with mathematical analysis.
In particular, we are interested in understanding how top-down processes control and configure computation in cortical circuits. Communication-through-coherence (CTC) is a concept which links selective processing to neural rhythms in the Gamma and Alpha/Beta frequency ranges. Here we propose a hierarchical model in which oscillations self-organize by lateral recurrent and feedforward interactions, thus allowing to gate visual signals in dependence on the current behavioral task (see Figure). More recently, we started investigating how a distributed, hierarchical system with different cortical areas specialized in processing different visual features is optimally configured for a given task. Our ultimate goal is to explain key physiological findings about attentional modulation in the visual system and psychophysical evidence for distributed functional configuration in a coherent model.
We study prototypical converging network-motifs of neuronal populations in the visual hierarchy, which are putatively involved in many visual attention tasks. Competing stimuli are processed in separate populations in lower visual areas, here V1, whereas information is integrated in downstream areas, here V4. Each population consists of excitatory and inhibitory neurons that due to local feedback can generate rhythmic activity upon external stimulation. Under minimal and physiologically consistent assumptions on intra- and interareal connectivity, we showed that these networks self-organize into states where an attended stimulus is preferentially processed in downstream populations (Figure adapted from Harnack D., Ernst U.A., Pawelzik K.R. A model for attentional information routing through coherence predicts biased competition and multistable perception. J. Neurophysiol. 114: 1593-1605, 2015).
Empirical studies have correlated selective information processing to different physiological signatures and neural mechanisms. But which of these mechanisms really implements a visual computation, and which observed signatures are just byproducts of a complex recurrent dynamics?
By combining experiments (Brain Research Institute, University Bremen), microsystems technology (IMSAS, University Bremen) and computational work we investigate how the brain selectively processes visual information. In particular, we use two different approaches to causally identify potential mechanisms, namely a) intra-cortical microstimulation to probe information transfer in dependence on the current cortical state, and b) tagging an external signal with a random flicker modulation which allows tracking visual information through the visual system's hierarchy. In this project funded by the DFG via the Schwerpunktsprogramm SPP 1665, our group performs data analysis and develops methods for online signal tracking in collaboration with experimentalists. Furthermore, we investigate paradigms for probing strongly recurrently coupled networks by ICMS, hereby using internal network states (e.g. bistable attractors) to push the system into a desired attractor and link its response to potential mechanisms of signal routing.
Sample computational model setup. The model consists of three distinct neural modules: X,Y and Z. Each module consists of excitatory (e) and inhibitory (i) point neurons with inter-areal connectivity parameters set up to establish realistic neural activity, containing 1/f noise as well as oscillation resonance in the desired frequency range (e.g. gamma). X and Y represent lower level populations (e.g. V1) with feedforward connections to a higher level population (e.g. V4). The lower populations are driven by Poisson drives modulated by distinct flicker signals, allowing us to track either stimulus information routing throughout the network by using spectral coherence. The resulting network activity demonstrates multi-stable states and selective attention via routing by synchronization -- Z enters a favorable phase relationship with either X or Y, its activity reflecting the respective stimulus. Further, we simulate intercortical microstimulation (ICMS) by inserting a current pulse into the desired target population. By delivering a precisely-timed current pulse during a specific state of the network's activity (e.g. at a specific phase of the ongoing oscillation), we can reliably force the network to switch to or maintain a desired stable state (e.g. force Z to maintain a favorable phase relationship with X, routing stimulus X information and suppressing stimulus Y).