"The artificial brain" IBM TrueNorth will learn from the real brains

IBM announced the first call that throws to its future TrueNorth computer chips, the so-called "neuromorphic" computers, which are physically structured as a brain: to learn a lot of data collected by real human brains and to extract meaning from them. Such tasks for most traditional computers are almost impossible, but IBM hopes that with its new technology will have the opportunity to observe the human brain in real time - and perhaps interfere.

Particularly large and very dangerous epileptic seizures are the result of large-scale storms of electrical activity that rage in the two hemispheres of the brain. Doctors hardly find reliable signatures of these storms, so it is often not possible to detect them in advance and apply effective countermeasures. But neural networks are ideal for finding such complex models. IBM feeds its new neuromorphic TrueNorth chips with EEG data from epileptics in the hope that they will be able to identify the patterns associated with major seizures. However, the real potential lies in the application and unique advantages of the neuromorphic architecture.

IBM has once linked ordinary digital computers to the simulation of the human brain. It was just as complex, but it worked 1500 times slower than the original. IBM researchers noted that the real limiting factor is not the computing power, but the electricity. If we take the usual digital algorithms used to run a slow simulation, and just speed them up 1500 times, we need about 12 gigawatts of energy. And while laboratories are not equipped with nuclear reactors, we will have to find a better way.

Fortunately, we have a good drawing: the human brain. After all, the human brain works with the entire complexity of the human brain constantly and requires only 20 watts - enough to power a small light bulb. IBM TrueMorph neuromorphic chips also have increased energy efficiency.

These chips are perfect for running code analyzing data. A sophisticated and delicate analysis of data using traditional "iron" will lead to the discharge of a huge battery in a couple of hours; with TrueNorth or similar architecture, it may be possible to wear them on the body and maintain work for up to or even a week.

IBM's ultimate goal is to use TrueNorth to detect seizures in real time - using an implant or a wearable device. An energy efficient chip will be able to observe and read the user's EEG, identifying an approaching attack. Then he will be able to give you a signal through a smartphone or immediately to a health center, and in the future even to introduce the necessary medications.

Wearable technologies require revolution either in the energy sphere or in energy efficiency. If we are ever going to acquire devices that receive electricity from your clothes, for example, we will need chips that do not require large amounts of energy. Neuromorphic chips may become one of these solutions. Monitoring epilepsy is just one of the possible uses.

The article is based on materials https://hi-news.ru/computers/iskusstvennyj-mozg-ibm-truenorth-budet-uchitsya-u-nastoyashhix-mozgov.html.