November 13, 2017
In an interview, the U.S. National Energy Research Scientific Computing Center’s (NERSC) Prabhat discusses deep learning, machine learning, and the challenges of applying them to science.
“I think of deep learning . . . as a subset of machine learning, which in turn is closely related to the field of statistics, and all of them have to do with solving inference problems of one kind or another,” Prabhat says. He cites the accessibility of big data, more powerful computers, and their convergence as driving deep learning forward, and notes the preponderance of convolutional network and long short-term memory architectures in deep-learning scientific applications.
Prabhat says NERSC has about 70 users employing deep-learning software, and it has identified and deployed several popular deep-learning frameworks on its Cori system. The challenges Prabhat foresees for deep learning include handling complex and diverse datasets, performance and scaling, and addressing “a lack of theory, interpretability, uncertainty quantification, and the need for a formal protocol.”
View Full Article
Abstracts Copyright © 2017 Information Inc., Bethesda, Maryland, USA
No entries found