Pages Navigation Menu

SHOWFUN - Show & Fun & More!

IBM’s TrueNorth artificial brain to watch, learn from real brains

IBM has announced an exciting first challenge for its upcoming TrueNorth computer chips, so-called “neuromorphic” computers physically structured like the brain: Look at a mass of data collected from real human brains, and make sense of it. That sort of pattern finding is something conventional computers are bad at doing, but IBM hopes with new technology it will be able to watch the human brain in real-time — and potentially intervene.

Big, truly dangerous epileptic seizures are a result of large-scale storms of electrical activity raging back and forth between the two hemispheres of the brain. Doctors have had a hard time finding reliable signatures of these storms that can be detected early enough to allow effective counter-measures, but neural networks are perfectly suited to finding such complex patterns. IBM is feeding its new, neurally inspired TrueNorth chips reams of electro-encepholagram (EEG) readings from epileptics in the hope that they can find patterns associated with major seizures. But the real potential lies in application — and the unique advantages of neuromorphic architecture.

IBM once networked regular digital computers together into a simulation of a human brain. It had the full level of complexity, but ran about 1,500 times slower than the real thing. The researchers at IBM pointed out the real limiting factor is not computational power, but electrical power. Because if we took the regular digital algorithms used to run the slow simulation, and simply ran it 1,500 times faster, we would end up using something like 12 gigawatts of power to do so. If you don’t feel like ringing your lab in nuclear power plants, you’ll have to come up with something significantly better.

Thankfully, we have a blueprint: the human brain. After all, a human brain runs the complexity of a human brain in real time, and it does so for as little as 20 watts, or enough to power a small lightbulb. IBM’s neuromorphic chips, called TrueNorth, have a portion of that increase in energy efficiency.

What that means is that the chips best suited to running data-mining code, those that are physically structured like a neural network, are also the chips with the best practical ability to apply those programs in the real world. Complex, always-on data mining hardware could run down even an enormous battery in just a couple of hours using conventional hardware; with TrueNorth or a similar architecture, it might be possible to stay on our bodies and keep learning all day, or even all week.

The ultimate goal is to use TrueNorth to detect seizures in real-time through an implant or wearable solution. A chip, power-efficient enough to last, could monitor and sift EEG readings from a wearer to identify an oncoming seizure. It can then alert you via your smartphone or, hopefully, one-day contact medical services directly or even administer medication if needed.

Wearable tech needs either a power revolution, or a power efficiency revolution. If we’re ever going to power real devices with electricity harvested from your clothes, for instance, we’re going to need chips that can run on such paltry amounts of power. Neuromorphic chips could be one such technology. Monitoring epilepsy is just the beginning of the applications.

Leave a Comment

Captcha image