• Subscribe

The human brain: a recurrent network of networks

As well as having the potential to revolutionize our understanding of the brain, the Human Brain Project will also build new platforms for 'neuromorphic computing' and 'neurorobotics', allowing researchers to develop new computing systems and robots based on the architecture and circuitry of the brain. Video courtesy the Human Brain Project on Vimeo.

Scientists from across the globe met in Switzerland last month for the launch of the Human Brain Project. Billed as the world's most ambitious neurosciencfle project, it seeks to develop methods that will enable a deep understanding of how the human brain operates. The Human Brain Project, coordinated by Henry Markram of Ecole Polytechnique Fédérale de Lausanne in Switzerland, has 135 partner institutions and is co-funded by the EU with an estimated budget of €1.2 billion. The knowledge gained through the project is expected to be a key element in developing new medical and information technologies.

Earlier this year, during a presentation at ISC '13, Markus Diesmann from Germany's Jülich Research Centre (FZJ) provided some insights into the complexities surrounding this epic undertaking. He spoke about his work on simulating a circuit of the whole-brain network and discussed how specialized software can help represent the neurons and synapses of the full circuit.

Diesmann is working on simulating the chattering that occurs between the 86 billion interconnected neurons. In each cubic millimeter of brain tissue there are already around 100,000 neurons, and each one makes contact with around 10,000 other neurons. This means there are around 1,000,000,000 contact points (synaptic junctions) that allow neurotransmitters to pass to another cell.

"The challenge is to organize the computer memory in such a way that one can represent all these contact points," explains Diesmann. "Each nerve cell receives 50% of contacts from inside this cubic millimeter, but the other 50% come from outside their local network." He continues: "This means if you want to make useful statements about the activity in the brain you also have to look at a larger area. As each neuron is interacting with 10,000 others, a simulation on a supercomputer will have the neurons distributed over different compute nodes, and network communication must be highly optimized."

Diesmann is fortunate to have access to Europe's second fastest supercomputer, JUQUEEN, a 5 petaFLOPS system housed at FZJ, and has also used Japan's K supercomputer, which in late 2011 became the first computer to top 10 petaFLOPS. "With K one can roughly generate 109 neurons, but the human brain has 1011 neurons, so with present technology you need a computer 100 times larger than K," says Diesmann. "The brain is of course more complex than our models and nobody knows how much detail we will have to put into the model and which level of description is sufficient." He adds: "If you assume this level of description, we require an exascale system and we are only two orders of magnitude away".

However, the architecture of a supercomputer is still considerably different to a natural neural network because the brain's wiring is in three dimensions. Compare the 10,000 contact points of a neuron with the three contact points of a basic electronic transistor. "There is a huge problem in getting enough wiring on these boards to represent a biological neural network," says Diesmann. "No hardware technology presently squeezes enough cable into the volume, and this is one of the key technical barriers that we have today." Only software simulations can map the 3D connectivity of the brain on to the existing standard computer technology hardware.

NEST (NEural Simulation Tool) is open-source software that can represent many neurons and their contact points. It focuses on the dynamics of neural systems, rather than on the exact morphology of individual neurons. A computational neuroscientist can represent mathematical models of nerve cells, synapses, and network structure in the computer and the software takes care of efficiently parallelizing the simulation on laptops, high performance clusters, or supercomputers. At the technical level there are two stages: the creation of the network and the simulation. Neurons in the central nervous system generate voltage spikes called action potentials, with the timing of these pulses transmitted to the other neurons via the synapses. NEST is ideal for building networks of these spiking neurons.

The number of NEST publications has exploded in the last five years. The NEST Initiative was recently set up as researchers become increasingly interested in full-scale networks, rather than concentrating on individual neurons or small circuits. The software still cited in the majority of publications though is NEURON, which focuses on the detailed description of individual neurons and small networks. NEST is efficient at memory usage and has clever communication strategies for shipping the activity data from one node to another. The idea is that you can solve a few neurons in detail in NEURON, then analyze a larger network using NEST. The simulation engines can also be coupled for multi-scale simulations by MUSIC, which allows simulators to exchange data during run-time. Using these software tools, Diesmann recently co-authored a paper showing that the differences in firing activity in different cell types can be understood on the basis of anatomy.

Join the conversation

Do you have story ideas or something to contribute? Let us know!

Copyright © 2020 Science Node ™  |  Privacy Notice  |  Sitemap

Disclaimer: While Science Node ™ does its best to provide complete and up-to-date information, it does not warrant that the information is error-free and disclaims all liability with respect to results from the use of the information.

Republish

We encourage you to republish this article online and in print, it’s free under our creative commons attribution license, but please follow some simple guidelines:
  1. You have to credit our authors.
  2. You have to credit ScienceNode.org — where possible include our logo with a link back to the original article.
  3. You can simply run the first few lines of the article and then add: “Read the full article on ScienceNode.org” containing a link back to the original article.
  4. The easiest way to get the article on your site is to embed the code below.