Researchers will have the computing power to simulate the neural network of the entire human brain.

Researchers at the Jülich Research Center have made major breakthroughs in simulating neural connections in the human brain. They use the NEST open source neural simulation tool to have the computing power of 100 billion neurons that mimic the human brain on supercomputers.

A team of international scientists invented an algorithm that represents an important step forward in the study of neural connections in the human brain. The new algorithm is described in an open access paper published in Frontiers in Neuroinformatics to allow the simulation of 100 billion interconnected neurons in the human brain on supercomputers. This work was done in collaboration with researchers at the Jülich Research Center, the Norwegian University of Life Sciences, the University of Aachen, RIKEN and the KTH Royal Institute of Technology.

Open source neural simulation tool

The algorithm was developed using NEST* ("Neural Simulation Tool"), an open source simulation software that is widely used in the neuroscience community and is the core simulator for the European Human Brain Project. The researchers explained in an announcement that with NEST, the behavior of each neuron in the network is represented by a small number of mathematical equations.

According to Markus Diesmann, director of the Jülich Institute of Neuroscience and Medicine, since 2014, large-scale nerves using NEST on the petascale ** K supercomputer on the RIKEN and JUQUEEN supercomputers at the Jülich Supercomputing Center in Germany Network simulation can simulate the connection of about 1% of neurons in the human brain. These simulations used previous versions of the NEST algorithm.

Why can't a supercomputer simulate the whole brain or why not? Susanne Kunkel, senior writer at KTH Royal Institute of Technology in Stockholm, explained: "Before the neuron network simulation, neurons and their connections need to be created virtually."

In the simulation process, the action potential (short electrical pulse) of the neuron needs to be sent to all about 100,000 small computers called nodes, each of which is equipped with multiple processors that perform actual calculations, and then each The node checks which of these pulses is associated with the virtual neurons present on the node.

This process requires one bit of information per processor per neuron in the entire network. For a network of one billion neurons, most of the memory of each node is consumed by a single message for each neuron. Of course, the amount of computer memory required by each processor required for these extra bits of each neuron increases with the size of the neural network. To exceed 1% and simulate the entire human brain, the available memory per processor is 100 times larger than the current supercomputer.

Brain simulation software running on current gigabit supercomputers can only represent about 1% of neuronal connections in the cerebral cortex (dark red areas on the left). In the next generation of billion-dollar supercomputers, it will be possible to simulate 10% of the entire human brain's neuron connections (centers), which is 10 to 100 times better than today's high-end supercomputers. However, using the same amount of computer memory as the current supercomputer, a new algorithm can simulate a 100% human brain (full brain simulation) on a tens of billions of supercomputers.

With the control of memory consumption, simulation speed will become the main focus. For example, a large simulation of 520 million neurons connected by 5.8 trillion synapses on Jülich's supercomputer JUQUEEN takes 28.5 minutes to calculate one second of biological time. The researchers calculated that with the improved algorithm, the time would be reduced to only 5.2 minutes.

"The combination of tens of billions of computational speed hardware and [coming NEST] software brings together fundamental aspects of brain function, such as plasticity and learning, which are unfolding within minutes of biological time," Diesmann said. Within the scope of our research.

The researchers found that the new algorithm will also make the simulation of currently available petascale supercomputers faster.

NEST simulation software update

In the next simulation software release released by the Neural Simulation Technology Initiative, researchers will provide new open source code to community researchers for free.

For the first time, researchers will have the computing power to simulate neural networks that scale the entire human brain.

Kenji Doya of the Okinawa Institute of Science and Technology (OIST) was probably one of the first to try it. He said, "We have been using NEST on K computers to simulate the complex dynamics of the basal ganglia loop in health and Parkinson's disease. We are very happy to hear about the new generation of NEST, which will enable us to work on the K computer. Run a whole brain simulation to clarify the neural mechanisms of motor control and mental function."

Note:

* NEST is a simulator for neural network models that focuses on the dynamics, size and structure of the nervous system, rather than the exact shape of a single neuron. NEST is suitable for peak neuron networks of any size, such as information processing models such as mammalian visual or auditory cortex, network activity dynamics models (such as layered cortical networks or balanced random networks), and learning and plasticity models.

** The Petascale supercomputer runs at a rate of teraflops per second (1015 floating point operations per second). Future E-class supercomputers will run at speeds of 1018 flops per second (1018 flops/s). At present, the fastest supercomputer is Sunway TaihuLight of China's Wuxi National Supercomputing Center, which runs at a speed of 9.3 trillion times per second.

*** At Jülich, this work was supported by the Neuroscience of the Simulation Laboratory, the Bernstein Network Computing Neuroscience facility at the Jülich Supercomputing Center. Part of the funding comes from the EU's Human Brain Project (HBP) and the EU Horizon 2020 research and innovation program, as well as the exploratory challenge to the post-K computer (the neural mechanism of understanding ideas and its application in artificial intelligence) Ministry of Science (MEXT). Through their joint project between Japan and Europe, researchers hope to contribute to the establishment of the International Brain Initiative (IBI).