Machine and Deep Learning on Power Systems

Abstract to be presented at Summit 2016

Machine and Deep learning play an increasingly important role in the exploitation of IT infrastructure. Machine Learning capabilities represent are used to power internet search, rank search results, identify top articles for news headlines, and so forth. At the same time, recent advances in Deep Learning have made deep learning the most successful technology for image recognition, speech processing, and text classification.

As human ingenuity to identify rule-based patterns and programming capacity to implement them as models of human cognition reach the limits of what can be implemented in rule and heuristic based systems, recent advances in Deep Learning have filled this void, benefitting from the availability of burgeoning compute capacity in modern multicore systems and compute accelerators such as GPUs and FPGAs.

With OpenPOWER technologies ranging from CAPI to NVlink to high-bandwidth datacenter network-connected clusters, OpenPOWER is rapidly becoming the natural platform to drive these technologies further. In this talk, I will describe the available machine and deep learning frameworks for Power, as well as present end to end solutions bringing together clustered GPU-based neural network training, FPGA-accelerated classification and CPU-centered enterprise workloads in a single solution.

 

Speaker Bio – Dr. Ruchir Puri for Dr. Gschwind

Dr. Gschwind is Chief Engineer for Machine Learning and Deep Learning, and Chief Architect and Senior Menager of Power Systems Architecture in IBM Systems Group.  In addition, Dr. Gschwind chairs the OpenPOWER Hardware Architecture Work Group.   In this role, he is responsible for the Power Architecture roadmap and the ongoing evolution and definition of the Power Architecture.  Dr. Gschwind is an IBM Master Inventor, a Member of the IBM Academy of Technology, a member of the ACM SIGMICRO Executive Board and a Fellow of the IEEE.