Sign In

Communications of the ACM

ACM TechNews

Eagle-Eyed Machine Learning Algorithm Outdoes Human Experts


Defects in radiation-damaged materials resemble a cratered lunar surface.

Researchers from the University of Wisconsin-Madison and Oak Ridge National Laboratory have taught computers to spot microscopic radiation damage to materials being considered for nuclear reactors.

Credit: Kevin Fields

Collaborators from the University of Wisconsin-Madison (UW-Madison) and Oak Ridge National Laboratory (ORNL) have taught computers to quickly and consistently spot and analyze microscopic radiation damage to materials being considered for nuclear reactors.

UW-Madison's Wei Li and Dane Morgan, with ORNL's Kevin Field, developed a machine learning algorithm to make a neural network capable of rapidly mining electron microscopy images of radiation-exposed materials to identify a specific type of damage, known as dislocation loops.

Following training with 270 images, the neural network, in conjunction with a cascade object detector, correctly detected and classified about 86% percent of dislocation loops in a series of test images, versus 80% found by human experts.

From University of Wisconsin-Madison News
View Full Article

 

Abstracts Copyright © 2018 Information Inc., Bethesda, Maryland, USA


 

No entries found