From Structure to function: sensory processing and decision in full-scale cortical network models
In our daily life we permanently and seamlessly perceive our environment focusing on and evaluating the information relevant to us, on the basis of which we take decisions and control our behavior. In the primate brain, this process requires the coordinated action of different brain areas where each area subserves specific functions.
Experimental approaches to primate decision making involve complex behavioral experiments that require the subject to achieve a cognitive task according to pre-set rules, much like in a video game. In monkeys such behavioral experiments are combined with invasive measurement of neural activity. Over the past decades these experiments have led to a range of insights into how task-relevant information is represented at different cortical processing stages, from early sensory processing to descending motor control (e.g. Bichot & Schall, 1999; Romo et al., 1999; Kim & Shadlen, 1999; Pesaran et al., 2008; Rickert et al., 2009; Stanford et al., 2010; Curtis & Lee, 2010; Harvey et al., 2012).
However, state-of-the-art experimental techniques of parallel neuron recordings only allow monitoring 10s to a few 100s of neurons, only providing a glimpse of the overall network activity and computation.For a deeper and comprehensive understanding of cortical processing underlying cognitive functions that involve working memory and/or decision making we require novel modeling approaches that integrate anatomical and physiological knowledge at an appropriate level of complexity. Current models of decision making are simplified to mathematical abstract levels and we are thus far from an understanding at the cellular and network levels.
To date large-scale cortical network models are typically concerned with reproducing the dynamics of resting-state neural activity rather than generating function during task performance, and individual cortical areas tend to be represented in a highly simplified manner. While advances have recently been made toward more detailed modeling of both the local circuits and the corticocortical connections, these advances have not yet been geared toward describing function. Models that describe function, on the other hand, are typically of rather abstract nature. Also recently, attractor models have become a popular model for working memory and decision making (Wang, 2002; Deco& Rolls, 2006; Deco et al., 2010) and clustered networks of neurons in local circuits (Song et al., 2005; Lee et al., 2016; Litwin-Kumar & Doiron, 2012; Rost et al., 2017) have been suggested as substrate for dynamical attractors in the neocortical network.
In this project we propose to bring together these strands, combining clustered topology of attractor networks with laminar organization in a full-scale spiking model of single and interacting cortical microcircuits with realistic connection probabilities. This will enable us to study how specific aspects of known anatomy affect and subserve cortical decision making processes.