Bagrat Amirikian, Principal Investigator

Three-Dimensional Lattice Model of the Motor Cortex: Emergence of Topologically Ordered Neuronal Structures with Coherent Properties

The long-term goal of this research project is to elucidate a relationship between the structure and function of fundamental distinct areas in the neocortex, in general, and the motor cortex, in particular, by a combination of theoretical methods with experimental approaches. The researchers are working to advance the understanding of whether and how the spatio-structural constraints on intrinsic connectivity affect the segregation of neurons into functional modules. They are working on a three-dimensional lattice model that allows for a fundamentally novel approach to studying directional operations performed in the motor cortex by providing means for explicit exploration of the link between the underlying local cortical structure and global collective properties of interacting cells that are substrates of this structure. This three-dimensional lattice model is heavily based on the accumulated knowledge of the neuroanatomy and neurophysiology of the motor cortex. The model will allow the researchers to bridge theoretical frameworks and experimental data in the domain of very large-scale simulations of networks of simplified neurons.

Recent work on this project has continued to focus on the use of neural network models to understand the processes underlying the brain’s control of movement. The researchers have developed a parallelized learning algorithm that was used to optimize an artificial neural network for the task of translating neural signals generated by populations of cells in the motor cortex into meaningful movement commands. They have tested the parallelized learning algorithm against serial versions and have found that it significantly reduces the amount of time required for the artificial neural network to converge on a near-optimal solution.

The researchers have also begun construction of a large-scale model of a neural network with realistic connectivity. Supercomputing resources have been used to benchmark the processing times required for serial simulations of the network. The group’s next task is to compare these results to those of a parallel version of the model, which is under development.