Attention

Selection of information that is relevant for the current behavioural goal is an indispensable feature of any system which is overwhelmed by a flood of input signals received at any moment in time. Attention is a concept?/mechanism? by which the brain focuses its computational resources on particular aspects of the current stimulus for enhancing information processing. In collaboration with the lab of Andreas Kreiter, we investigate neural signatures and mechanisms of attention in the visual system by performing data analysis on multi-electrode data and neural network simulations.

In particular, our research focuses on the role of oscillatory activity as an attentional gating mechanism: is it possible to selectively route the information flow between visual areas by synchronizing the corresponding neural populations with an appropriate phase lag? What is the corresponding architecture and dynamics of the networks involved in attentional gating? Does synchronization act as a 'carrier' for processing and transmitting visual information from attended regions of the visual field?

Information Routing by Attention

Attending a visual stimulus is accompanied by large increases in neural activity in the Gamma band (30-90 Hz) in visual areas V1 and V4. What is the functional relevance of this oscillatory activity for information processing in the visual system?

One possibility is that synchronization might serve to increase discriminability between different visual stimuli. We discovered this effect in local field potentials (LFPs), where attention improved the classification performance of visual stimuli considerably. Currently, we study potential neural mechanisms supporting this phenomenon. Our basic idea is that synchronization of randomly coupled integrate-and-fire neurons serves to amplify differences in the particular input patterns these neurons receive from visual stimulation.

A second hypothesis is that Gamma oscillations serve as a gating mechanism which dynamically routes information between different visual areas. This idea is investigated by combining experimental and modelling studies. The experiments are performed in the group of Prof. A. Kreiter. Their goal is to quantify how much information about independent luminance changes in two visual stimuli is conveyed to V4 neurons, in dependence on whether attention is placed on the one or on the other stimulus. Modeling comprises simulations of neural populations, whose activity is modulated by Gamma oscillations. Our aim is to identify putative mechanisms and network architectures that both implement gating, and reproduce neural data obtained from the experiments.

Classification of Cognitive States

Attending a location in the visual field is known to strongly modulate neural responses to visual stimuli. Can we use these signatures of attention to read the current cognitive state from recorded brain activity?

In collaboration with Prof. A. Kreiter, we investigate neuronal signatures of attention in multi-electrode local field potential (LFP) recordings from awake behaving macaque monkeys. By using methods from machine learning, ranging from linear classifiers to support vector machines, we predict the current stimulus and the current cognitive state from the brain signals. In particular, we identify specific signal features that carry information about the stimulus, and of the direction of attention towards specific locations in the visual field. Using time-resolved prediction techniques, we also seek to quantify how attentional load depends on task demands.

The high classification performance achieved in this project (>99% correct for classifying the attended location) bears a high potential for future brain-computer-interfaces based on signals recorded in the visual system (see e.g. Kalomed).