SC16 Salt Lake City, UT

Distributed Machine Intelligence Using Tensorflow


Authors: Dr. Karan Bhatia (Google)

BP Abstract: TensorFlow is the second-generation machine learning system from the Google Brain team, using tensor notation to describe neural networks as stateful dataflow graphs in ways that are both concise and flexible. This BOF session brings together developers of TensorFlow, users of TensorFlow, and members of the HPC community curious about how to develop and exploit deep learning systems for data analysis and autonomous operation.

Long Description: TensorFlow is the second-generation machine learning system from the Google Brain team, using tensor notation to describe neural networks as stateful dataflow graphs in ways that are both concise and flexible. Since the release of the first reference implementation in November 2015, TensorFlow has become the basis of hundreds of research and commercial systems. While the initial open-source reference implementation supported only a single node, a distributed runtime was added in version 0.8, and ports and re-implementations of TensorFlow are running on large parallel systems and custom processors. This BOF session brings together developers of TensorFlow, users of TensorFlow, and members of the HPC community curious about how to develop and exploit deep learning systems for data analysis and autonomous operation. The session will invite a technical expert from the TensorFlow team to provide a technical overview of the system, and a 4-5 invited panelists will discuss applications along with technical challenges. After the directed questions from the moderator, the audience will have an opportunity to ask questions to the experts. A report with the presentations and QA will be produced.


Birds of a Feather Index