Large-scale simulation of the visual cortex
Group leader: Prof. Dr. Markus Diesmann
In this project, we develop and simulate a spiking neural network model of multiple interacting areas of macaque visual cortex, connected in a layer-specific fashion. The model is implemented in NEST, and adapts and extends a local cortical microcircuit model developed by Dr. Tobias Potjans. This work contributes to BrainScaleS, a European FP7 project focused on understanding brain processing at multiple spatial and temporal scales.
This project addresses several issues:
1. In local cortical models, a large percentage of synapses remains unaccounted for. By including both local and remote connections, our model accounts for most synapses impinging on the neurons considered. It thus provides a more detailed and self-consistent description of the processes underlying the cortical dynamics.
2. The model includes millions of neurons and billions of synapses. Simulating systems of such scale requires massively parallel computation, for which we are uniquely positioned with access to the JUQUEEN and K supercomputers. In collaboration with the NEST Initiative, the project enhances knowledge on the efficient implementation of such large-scale models in terms of speed and memory consumption.
3. Besides implementation on supercomputers, versions of the model are being ported to neuromorphic and other dedicated hardware that is developed by collaborating groups as a major component of the BrainScaleS project. To this end, the local model has been reimplemented in the simulator-independent language PyNN. This part of the project contributes to expertise on dedicated hardware systems for neural network simulation.