Researchers at the University of Sherbrooke in Canada have outfitted a micro-electromechanical system (MEMS) with artificial intelligence for the first time, enabling neuromorphic computing in a microscale device and bringing edge computing a step closer.
Sherbrooke's Julien Sylvestre says "reservoir computing" based on time-dependent input is used to drive the dynamical system.
This has multiple degrees of freedom for responding to the input, which means the input is "mapped" along a pathway in a high-dimensional space. Each dimension corresponds to one of the degrees of freedom and creates various transformations of the input.
Sylvestre says the reservoir computing is trained via linear combination of the dimensions to yield the desired output for a given input.
The MEMS device serves as the dynamical system, using nonlinear dynamics of a thin silicon beam's spatial oscillation and producing a neural network to convert the input signal into the higher-dimensional space required for neural network computing.
From IEEE Spectrum
View Full Article
Abstracts Copyright © 2018 Information Inc., Bethesda, Maryland, USA
No entries found