U.S. Department of Energy

Pacific Northwest National Laboratory

TensorFlow for the extreme scale

‘Deep Learning’ applications are everywhere, and so too are tools that support it. Google recently open-sourced deep learning with TensorFlow, a program featuring which features algorithmic optimization support for single node multi-core systems and GPU systems. However, this is limiting, given the ever-increasing data volume of most applications. A team of PNNL researchers has now extended TensorFlow to run on extreme scale systems. With a planned release this spring, users can solve very large problems on supercomputers and cloud systems. PNNL will also have a release integrated with Machine Learning Toolkit for Extreme Scale (MaTEx), a collection of high performance parallel machine learning and data mining algorithms is planned for late spring. Learn more at http://goo.gl/lOucUQ.

| Pacific Northwest National Laboratory