Michigan supercomputer attempts to mimic brain of a cat

A cat's brain is the model for a new supercomputer developed by the University of Michigan and funded by DARPA.
Written by Andrew Nusca, Contributor

The brain of a cat is the model for a new supercomputer developed by the University of Michigan.

UMich computer engineer and assistant professor Wei Lu is working to develop a type of machine that can learn, recognize, and make more complex decisions and perform more tasks at once than conventional computers can.

The reason: a cat can recognize a face faster, and more efficiently, than today's supercomputer.

Today's most sophisticated supercomputer -- a massive system with more than 140,000 central processing units -- can accomplish certain tasks with functionality on par with a cat.

But it still performs 83 times slower than a cat's brain, Lu says.

Here's how a cat's brain works: neurons are interconnected by synapses, which act as reconfigurable switches that form pathways linking thousands of neurons. Synapses can recall those pathways based on strength and timing of the neurons' electrical signals.

In contrast, a computer's logic and memory functions are located at different parts of the circuit, Lu says. Because each unit is only connected to a few around it, rather than length pathways, conventional computers execute code line by line.

Computers excel at performing simple operations with limited variables.

Brains excel at performing many tasks at once in parallel.

That's why mammals can recognize faces far more quickly than computers.

Lu's work centers around a "memristor," a transistor device that acts like a biological synapse by remembering past voltages.

Lu demonstrated that a memristor can connect conventional circuits in a way that provides the very basis for memory and learning.

"We are building a computer in the same way that nature builds a brain," Lu said in a statement. "The idea is to use a completely different paradigm compared to conventional computers."

Lu has thus far connected two electronic circuits with one memristor and demonstrated that the system is capable of memory and learning, in the form of a process called "spike timing dependent plasticity."

"We show that we can use voltage timing to gradually increase or decrease the electrical conductance in this memristor-based system," Lu said. "In our brains, similar changes in synapse conductance essentially give rise to long term memory."

The next step? A larger system, of course.

That kind of advancement would compact supercomputer strength into a portable container.

Lu's got some serious financial backing, too. The Pentagon's Defense Advanced Research Projects Agency -- DARPA -- is funding his research, along with the National Science Foundation.

His research is published in the April edition of the journal Nano Letters.

Related on SmartPlanet:

This post was originally published on Smartplanet.com

Editorial standards