DeepSouth, the first supercomputer capable of simulating neural networks on the scale of the human brain, developed by Australian researchers, will be operational in 2024. It promises advances not only in the field of computing (especially AI), but also in neuroscience.
2024 will mark a turning point in neuroscience and computer science research with the launch of DeepSouth,
supercomputer developed by researchers from the International Center for Neuromorphic Systems (ICNS) at Western Sydney University (Australia).
It is capable of mimicking brain processes and will enable advances in the understanding of efficacy Artificial Intelligence (AI). The project is the result of an international collaboration and promises advances in fields ranging from robotics to health, while promoting more environmentally friendly computing.
The supercomputer is aptly named DeepSouth, a tribute to IBM’s TrueNorth system, which aimed to produce machines simulating large networks spiky and Deep Blue, the first computer to become a world chess champion. The name is also a nod to its geographical location.
Neuromorphism and energy efficiency
DeepSouth represents a significant technological breakthrough, especially thanks to its neuromorphic architecture, directly inspired by the biological processes of the human brain.
This innovative approach allows the supercomputer to simulate synaptic operations – interactions between neurons – on an unprecedented scale. With its ability to reach 228 trillion (228 x 1012) of synaptic operations per second, DeepSouth achieves performance levels that rival human neuronal activity.
This ability to reproduce the complexity and speed of neuronal interactions opens new perspectives in understanding the functioning of the brain and in the simulation of cognitive processes.
The most revolutionary aspect of DeepSouth is its energy efficiency. Unlike traditional supercomputers, which require large amounts of power to operate, DeepSouth manages to perform these massive synaptic operations while using much less power.
This performance is made possible by the use of high-end computer hardware, in particular Field Programmable Gate Arrays (FPGAs), which allow flexible post-production reconfiguration and thus energy optimization.
The Director of ICNS, Professor André van Schaik, declares va communicated : ” Progress in our understanding of how the brain computes with neurons is hampered by our inability to simulate large-scale brain-like networks. Simulating neural networks on standard computers using graphics processing units (GPUs) and multi-core central processing units (CPUs) is simply too slow and power-hungry. Our system will change that “.
Multidisciplinary application
In biomedicine, DeepSouth’s ability to simulate the human brain opens the door to deeper understanding
neurodegenerative diseases such as Alzheimer’s or Parkinson’s. Researchers will be able to use DeepSouth to model and study neuronal dysfunctions that could lead to new therapies or drugs.
In robotics, mimicking DeepSouth’s brain processes promises to make robots more autonomous and adaptive, able to process complex information with unprecedented speed and efficiency.
In addition, DeepSouth will certainly have an impact on space exploration and artificial intelligence. In space, energy management is essential. DeepSouth’s energy efficiency could therefore enable the development of more efficient navigation and data processing systems for space missions.
When it comes to artificial intelligence, DeepSouth offers the ability to create more sophisticated and, again, more importantly, less power-hungry machine learning algorithms. These algorithms could more closely mimic human thought processes, leading to advances in areas such as speech recognition, computer vision, and automated decision-making.
DeepSouth’s scalability is a major strength, allowing flexible adaptation to different research and application requirements. This feature allows users to add or remove hardware according to the specific needs of their projects, providing unparalleled customization.
This flexibility makes DeepSouth not only suitable for a wide range of research scenarios, but also economically viable for a variety of institutions and companies, from universities to startups.
In addition, as already mentioned, the key innovation is the reconfiguration of DeepSouth using FPGAs. FPGAs offer the ability to reprogram hardware after it has been manufactured, allowing new neural models to be integrated and rules to be learned over time. This ability to adapt and evolve is essential in such a dynamic field as neuroscience and artificial intelligence.
In addition, DeepSouth will be accessible remotely through a Python user interface, making it easy to use by researchers and developers without requiring extensive IT knowledge. This availability, combined with the system’s ability to mimic the processes of the human brain, will offer a solution for processing information closer to natural mechanisms.
The ICNS team at the University of Western Sydney collaborated with international partners, including Aachen University in Germany.
In addition, DeepSouth is set to play a leading role in developing Western Sydney as a cutting-edge technology hub. Its commissioning, planned for April 2024, represents an important milestone in the development of the region’s technological infrastructure.
DeepSouth’s impact will extend beyond academic boundaries, positively impacting the local economy, attracting investment, creating jobs and fostering innovation in key sectors such as artificial intelligence, robotics and biomedicine.