Machine Learning Algorithm Beats Human Experts At Detecting Microscopic Radiation Damage
Engineering researchers are quickly making it possible for the silicon brains to surpass human intelligence. In yet another excellent feat, a team of scientists from University of Wisconsin–Madison and Oak Ridge National Laboratory has been outranked by a machine learning algorithm in the detection and analysis of microscopic radiation affecting materials research used in nuclear reactors.
The current mechanisms for image analysis at the microscopic level often employ humans who might lack the expertise required to do the job. Human detection was found to be error-prone, inconsistent and inefficient.
This has long been a major research bottleneck. By quickly realizing that bringing artificial intelligence to materials science research, the researchers have been able to not just consistently analyze the images, but also much faster (in a fraction of seconds required for humans to do the job).
The team made use of statistical methods to guide the algorithm in improving its performance in image analysis. They made the computer rapidly screen a bunch of images of materials that had been exposed to radiation, and identify a specific type of damage.
Professor Dane Morgan and his collaborators taught a neural network to recognize a specific type of radiation damage, called dislocation loops.
After training with 270 images, the neural network and a machine learning algorithm called a cascade object detector could correctly identify and classify ~ 86% of the loops.
The research team believes that this research work is just the beginning and the future machine learning algorithms are going to leave all cyberinfrastructure researchers in awe.
Source and Image Credit: WISC