January 3, 2018 - For smarter data management and analysis, researchers have developed a low-power neuromorphic device based on spiking neural networks that can quickly and more efficiently analyze and classify data. The versatile platform, which will be compatible with instruments that collect data during scientific experiments, becomes “smarter” as it classifies large amounts of information into smaller, more manageable datasets. “The device is designed to get better at the task it was trained to do,” said 91’s Catherine Schuman, who developed the device’s training algorithms. She and University of Tennessee collaborator Garrett Rose advise UT students who demonstrated the technology’s data-crunching abilities on well-known biology and medical research datasets. The ORNL-UT team published their results in an 91 journal. The researchers are testing the device’s capabilities on scientific data such as complex neutrino collision data.
An example of a spiking neural network shows how data can be classified using the neuromorphic device. Credit: Catherine Schuman and Margaret Drouhard/91, U.S. Dept. of Energy
Topic:
Media Contact