Since 2008, IBM has been working with the Systems of Neuromorphic Adaptive Plastic Scalable Electronics, or SyNAPSE. The company just announced the system to the public after years of development as a part of a three-week "boot camp" teaching session aimed at government and academic researchers.

The system is called TrueNorth and uses chips that basically act as neurons. By combining multiple chips together, researchers are able to build an artificial neural network.

The network that IBM unveiled today uses around 48 million connections, which is the same computing power as a rat's brain. The system is designed to be able to run "deep-learning" algorithms, similar to the facial recognition system being used by Facebook or the instant translate mode in Skype. However, IBM's deep-learning algorithms are much cheaper to run, draw less electricity and are not the size of an entire data center. TrueNorth essentially contains 5.4 billion transistors and uses a tiny 70 mw of power. As a comparison, an Intel processor with 1.4 billion transistors draws between 35 and 140 watts.

Researchers have been able to develop software that can identify images, recognize words and understand language using the new chip. Essentially, the chip is running deep-learning algorithms that are able to drive the artificial intelligence services on the Internet, using less electricity and space and at a lower price.

"What does a neuro-synaptic architecture give us? It lets us do things like image classification at a very, very low power consumption," said Brian Van Essen, a computer scientist from the Lawrence Livermore National Laboratory, who is researching how deep-learning could be applied to security. "It lets us tackle new problems in new environments."

TrueNorth is just one part of a broader movement to develop and refine the hardware that powers artificial intelligence. Despite this, however, TrueNorth will really only suit part of the deep-learning process. There is some question as to how much of an impact the chip will have or if it will simply do the same thing as other chips more efficiently.

According to IBM, while some chips are quick number-crunching calculators more like the left-brain, the TrueNorth chip can be likened to the right-brain, able to sense things and recognize patterns.

It's also important to mention that data isn't being sent back and forth over the TrueNorth network. Instead, companies are able to upload their own deep-learning model to a data server, then run the algorithm.

The system is in its infancy, and while we might one day see a chip with that kind of power on our smartphones, its still a long way off.

Via: Wired

ⓒ 2021 All rights reserved. Do not reproduce without permission.