IBM researchers say they have built the digital equivalent of a rodent brain encompassing 48 TrueNorth chips, an experimental processor designed to emulate neurons.
IBM is holding a "boot camp" for government researchers and academics, where TrueNorth software is being developed. Some researchers have crafted software that can run deep-learning algorithms capable of recognizing spoken words, identifying images, and understanding natural language.
"The chip gives you a highly efficient way of executing neural networks," says University of Michigan professor Jason Mars.
TrueNorth chips promise to run algorithms in smaller spaces while consuming less electricity, and Lawrence Livermore National Laboratory's Brian Van Essen says this "lets us tackle new problems in new environments."
TrueNorth is currently suitable for only one aspect of deep learning--enabling the neural network to execute models it has been trained for--but IBM's Dharmendra Modha, a recipient of the ACM Gordon Bell Prize in 2009, notes this is appropriate. "We're trying to lay the foundation for significant change," he says.
The chip is equipped with 5.4 billion transistors, yet it draws only about 70 milliwatts of power. The smallness and low-power consumption of TrueNorth enables faster data processing as the information is not network-routed, and Mars thinks this will help embed more processing within devices.
View Full Article
Abstracts Copyright © 2015 Information Inc., Bethesda, Maryland, USA