Talk at SfN on prefrontal-parietal interactions

I’m giving at talk at the 2011 Society for Neuroscience meeting as part of a Nanosymposium titled ‘Neural Basis of Decision Making’, November 12. My talk is titled: Decision-making dynamics in a parietal-prefrontal loop model. My abstract is below:

Decision-making dynamics in a parietal-prefrontal loop model

Dept. of Neurobio., Yale Univ. Sch. of Med., New Haven, CT

Cognitive functions such as decision making and working memory involve a distributed interacting network of brain areas, with the prefrontal and parietal cortices at the core. However, the differential functions of these areas, and the role of interactions between them, are poorly understood. It is unclear, for example, whether the persistent activity observed during working memory originates locally within a brain area or instead arises through reciprocal loop interactions with other areas. Moreover, it is important to study how activity is coordinated in a distributed network during decision making and working memory, particularly if long-range projections are weaker than local recurrent connections.

We consider a model of two interacting modules, representing a prefrontal area and a parietal area, that is capable of both working memory and decision making. Each module is composed of excitatory populations selective for different choice options that compete with each other through mutual inhibition. By varying the strength of local recurrent excitation and inhibition, each module can operate in different regimes, e.g., exhibiting attractor dynamics or not. The selective populations send long-projections between modules onto excitatory and inhibitory cells. With this framework, we examine how the local and long-range connectivity in the network affects the dynamics of decision making.
We find that local competition dynamics requires strong connections within a module, which influences the module’s sensitivity to relatively weak long-range input from the other module. One module (e.g., prefrontal cortex) may need to control another module (e.g., parietal area LIP) to switch on or off a memory. We find that switching is possible even with weak long-range projections if they exhibit balanced excitation and inhibition.

A two-module network can display behaviors that are qualitatively distinct from those possible with a single module. The two modules can reach different choices (e.g., one module selects option A, whereas the other module selects option B) when both modules are strongly recurrent locally and long-range projections are relatively weak. These conflict states occur during difficult decisions, when evidence is noisy or even contradictory between the two modules. An external control signal can flexibly shift the network into and out of a regime that allows conflict states. We also study the integration of multiple sources of input in decision making (e.g., sensory evidence into one module and reward bias into the other module). We find that the combination of these inputs depends on the relative strengths of local and long-range connections.