Error message

  • Warning: include(/var/www/ failed to open stream: No such file or directory in include() (line 361 of /var/www/
  • Warning: include(/var/www/ failed to open stream: No such file or directory in include() (line 361 of /var/www/
  • Warning: include(): Failed opening '/var/www/' for inclusion (include_path='.:/usr/share/php') in include() (line 361 of /var/www/

Supercomputer simulates brain activity

A computer with 1 petabyte of main memory takes 40 minutes to calculate 1 second of activity in 1 percent of the brain… but this is nonetheless an extraordinary scientific success!

September 2014

Supercomputer IMALab News

How does the human brain work? Can a computer simulate human thought processes and discover how to learn? Scientists have been asking these questions for years, and science-fiction writers have gleefully explored the same subject.

A joint Japanese and German research group has recently carried out a project using the so-called K Computer (currently the fourth most powerful computer in the world) to simulate neural activity on a scale that has never been attempted before, in order to take some steps towards finding answers to these questions.

The K Computer uses neural network technology: this consists of mathematical models which represent the interconnection between elements described as artificial neurons… that’s to say, they imitate the properties and activities of living neurons and their connections.

Neural networks are used in neuroscience to create models that could potentially explain some aspects of human cognitive phenomena. For example, they are being employed in research into infant language-learning mechanisms, and into the hemispheric lateralization of letter recognition.

But how can we be sure that computer neural networks accurately represent the workings of the human brain, or at least accurately enough to contribute to our understanding of how it works?

The only way to find out is by comparing the results of neural networks with the results of human learning, and the more the two match up the more it can be generally accepted that the neural networks accurately represent the cognitive phenomenon they were built to study.

The project, a joint enterprise between Japanese research group RIKEN, the Okinawa Institute of Science and Technology Graduate University and the Forschungszentrum Jülich, an interdisciplinary research center based in Germany, represents the largest neuronal network simulation carried out to date.

The K Computer used NEST, the open-source Neural Simulation Technology tool, to replicate a network consisting of 1.73 billion nerve cells connected by 10.4 trillion synapses. To do this, the computer notched up a rated performance of 10.51 petaflops per second using 705,024 processing cores and 1 petabyte of main memory (1.048 million GB, which is roughly equal to the memory of 250,000 PCs).

This kind of computing power is practically impossible for us to grasp, and yet this mighty monster of artificial intelligence took 40 minutes to crunch data relating to just one second of brain activity… if we remember that the human brain contains 86 billion neurons, this means that the supercomputer only managed to model roughly 1% of the brain.

This still represents an extraordinary success which marks a milestone in our study of the human brain and also demonstrates just how majestically gigantic its complexity and richness really is.

As for supercomputers, if petascale computers like the K Computer are capable of representing one per cent of the network of a human brain today, then we know that simulating the whole brain at the level of its individual nerve cells and their synapses will be possible when we can use exascale computers. Some researchers have speculated that exascale computing may be achieved in about ten years’ time, some go so far as to estimate as soon as 2020.