Analog Deep Learning, New AI Branch Could Be Employed for More Efficient Self-Driving Cars, Fraud Detection, Medical Image Analysis

A research team from the Massachusetts Institute of Technology set out to push the speed limits of a type of human-made analog synapse they had developed before.

Specifically, as specified in a SciTechDaily report, the researchers employed a practical inorganic material in the process of fabrication that allowed their devices to run one million times quicker than the past versions, which is about one million times faster as well, compared to the synapses in the human brain.

Essentially, the amount of time, initiative, and money required to train ever-more-complicated neural network models is soaring as scientists push the limits of machine learning.

A new branch of artificial intelligence, known as deep analog learning, promises faster processing with only a portion of energy use.

'Neurons' and 'Synapses

Programmable resistors are the main building blocks in analog deep earning, just as transistors are the central elements for building digital processors.

By repeating arrays of programmable resistors in complex layers, researchers can develop a network of analog artificial "neurons" and "synapses" that perform computations like a digital neural network.

Such a network can be trained to attain complex AI tasks like neutral language processing and image recognition.

Moreover, this organic material makes the resistor exceptionally energy-efficient as well. Unlike materials in their device's earlier version, the new material is compatible with silicon fabrication approaches.

Such a change has allowed the fabrication of devices at the nanometer scale and could pave the way for incorporation into commercial computing hardware for deep-learning applications.

Programmable Resistors

According to Jesus del Alamo, the study's senior author and Donner Professor in MIT's Department of Electrical Engineering and Computer Science, with that key insight and very powerful techniques they have at MIT nano, they have been able to put such pieces together and show that such devices are intrinsically very quick and operate with quite "a reasonable voltage."

This work published in the Science journal explained that the senior author had put the device at a point where it now looks promising for future applications.

Additionally, the working mechanism of the device is the electrochemical insertion of the tiniest ion, the proton, into an insulating oxide to modulate its electronic conductivity.

Since they are working with extremely thin devices, senior author Bilge Yildiz, senior author of the study and the Breene M. Kerr Professor in the departments of Nuclear Science and Engineering and Materials Science and Engineering, added, they could fast-track the motion of the said ion by using a strong electric field and push the ionic devices "to the nanosecond operation regime."

The said programmable resistors drastically accelerate the speed at which a neural network is trained while considerably decreasing the cost and energy to conduct the training, a related Tech Xplore report specified.

The latest development could help researchers create deep learning models much more rapidly, which could be employed in uses such as self-driving cars, fraud detection, and medical image analysis.

Related information about synapses in the human brain is shown on New Sparky's YouTube video below:

Check out more news and information on Artificial Intelligence in Science Times.