Simulation of a Human-Scale Cerebellar Network Model on the K Computer

Area of Science:

• Computational Neuroscience
• Neuroscience
• Supercomputing

Background:

• Simulating the human brain at the individual neuron level is a key goal in computational neuroscience.
• The cerebellum, comprising 80% of the brain's neurons, presents a significant computational challenge.
• The K supercomputer offers substantial computational power for complex simulations.

Purpose of the Study:

• To construct and simulate a human-scale spiking network model of the cerebellum.
• To assess the feasibility of simulating the cerebellum on a flagship supercomputer.
• To benchmark the simulation's performance on a cerebellum-dependent task.

Main Methods:

• Developed a human-scale cerebellar model with 68 billion spiking neurons using the MONET (Millefeuille-like Organization NEural neTwork) simulator.
• Utilized the K supercomputer for the simulation.
• Performed simulations of the optokinetic response, a cerebellum-dependent eye movement task.
• Evaluated the simulator's scalability by varying the number of compute nodes from 1,024 to 82,944.

Main Results:

• Successfully reproduced plausible neuronal activity patterns for the optokinetic response, consistent with experimental observations.
• Demonstrated good weak-scaling properties of the MONET simulator for the cerebellar network model.
• Achieved the first-time simulation of a human-scale cerebellum using 82,944 nodes on the K computer, albeit with a slowdown factor of 578.

Conclusions:

• The K supercomputer, in conjunction with the MONET simulator, is capable of simulating a human-scale cerebellar model.
• This study represents a significant step towards large-scale brain simulations.
• The findings highlight the potential of supercomputing in advancing computational neuroscience research.