Free Inference and Instant Training: Breakthroughs and Implications

  • Vivienne Sze, Geoff Gordon, Matthai Philipose, Amar Phanishayee, Christopher Re, Michael Jordan | Carnegie Mellon
  • Research Faculty Summit 2018

The fact that many commonly used networks take hours to days for training has motivated recent research towards reducing training time. On the other hand networks, once trained, are heavyweight dense linear algebra computations, usually requiring expensive acceleration to execute in real time. However, recent advances in algorithms, hardware, and systems have broken through these barriers dramatically. Models that took days to train are now reported to be trainable in under an hour. Further, with model optimization techniques and emerging commodity silicon, these models can be executed on the edge or in the cloud at surprisingly low energy and dollar cost. This session will present the ideas and techniques underlying these breakthroughs and discuss the implications of this new regime of “free inference and instant training.”

Speaker Details
Vivienne Sze
Research Director Geoff Gordon
RESEARCHER Matthai Philipose
Matthai Philipose builds sensor-based systems that allow computers to understand and act on human state. He most recently co-led the Everyday Sensing and Perception (ESP) project at Intel Labs, which had the modest goal of understanding 90% of a person’s life at 90% accuracy using mobile sensing. He has a strong subsidiary interest in applying such systems to the long-term care of the elderly. To this end, he has collaborated with Intel product groups, universities and government organizations to build and field-test novel telecare systems. Matthai has a Ph.D. from the University of Washington and a B.S. from Cornell University.
RESEARCHER, MSR-Redmond Lab Amar Phanishayee
Amar Phanishayee is a Ph.D. candidate at Carnegie Mellon’s Computer Science Department. The goal of his research is to enable the creation of high-performance, efficient networked systems for large-scale data-intensive computing. His research so far has addressed problems across the distributed systems stack: from new hardware to techniques to use it efficiently; from network protocols to distributed systems & consistency protocols. Amar was awarded an IBM Research Fellowship (2009, 2010), a ThinkSwiss Research Scholarship, and a SOSP Best Paper Award in 2009.
Christopher Re
Michael Jordan

Series: Microsoft Research Faculty Summit