Real-time simulation of a cat-scale artificial cerebellum on PEZY-SC processors

Abstract

The cerebellum is a part of the brain that plays essential roles in real-time motor learning and control and even cognitive functions. Thanks to the large amount of anatomical and physiological data, we implemented a spiking network model of the cerebellum on Shoubu, an energy-efficient supercomputer with 1280 PEZY-SC processors at RIKEN. Our artificial cerebellum consists of more than one billion neurons, which is comparable to the whole cerebellum of a cat. Using 1008 of 1280 processors on Shoubu, we achieved real-time simulation, that is, a computer simulation of the model for 1 s completes within 1 s of the real world with temporal resolution of 1 ms. Effective performance was estimated as 68 teraflops in single-precision floating points, suggesting 2.6% of peak performance. We expect that the artificial cerebellum will be applicable to various engineering applications, such as robotics and brain-style artificial intelligence.