acm-header
Sign In

Communications of the ACM

ACM TechNews

AI Learns by Doing More with Less


Colorful illustration of human neuronal system

a tiny insect brain has less than a million neurons but has a diversity of behaviors and is more energy efficient than current AI systems. These tiny brains serve as models for computing systems that are becoming more sophisticated as billions of silicon

New research from Washington University in St. Louis finds that the ability of silicon neurons to learn to communicate and establish networks is key to producing artificial intelligence systems that are as energy-efficient as biological ones.

Researchers previously found that neurons in a computational system act as though they are embedded in a rubber sheet; the new study demonstrates how neurons learn to choose the most energy-efficient perturbations and wave patterns in the rubber sheet. Each neuron adjusts the electrical stiffness of the rubber sheet to vibrate the entire network in the most energy-efficient manner, using only local information.

A silicon neuron, researchers discovered, can test all communications routes at once and identify the most efficient way to connect to complete a specific task.

From Washington University in St. Louis
View Full Article

 

Abstracts Copyright © 2021 SmithBucklin, Washington, DC, USA


 

No entries found