AI Research at Scale
- Speaker: Drs M. Cornero & D. Grewe (Google Deep Mind)
- Host: Antoniu Pop
- 22nd January 2018 at 14:00 in KB L.T 1.3
DeepMind's mission is to solve intelligence. We have 100s of Machine Learning researchers working towards this goal, each running many computationally expensive experiments. In this talk, we discuss the computational challenges we're faced with to enable research at this scale. We illustrate elements of our holistic approach composed of large distributed heterogeneous systems, including CPUs, GPUs and custom accelerators such as Tensor Processing Units, as well as the associated programming models, compilers and tools.
BiographyDominik did his PhD at the University of Edinburgh where he worked on mapping OpenCL programs to heterogeneous systems. He then joined Google to work on the performance of Chrome on Android before transferring to DeepMind. At DeepMind, Dominik leads the performance team whose mission is to accelerate AI research by enabling researchers to scale their experiments to increasingly powerful computing systems.
Marco works at DeepMind since August 2017 as compiler engineer in the Performance team. Before that he worked at ARM for four years as Mali GPU compiler architect, where he developed compilers for new GPU architecture explorations as well as for production drivers. From 2007 to 2013 he was an ST-Ericsson Fellow in charge of Advanced Computing, and before that he worked at ST-Microelectronics as director of Advanced Computing. Marco holds a Ph.D. in "Behavioral Synthesis of VLSI Systems" (CAD/EDA) from the University of Genova, Italy, in cooperation with IMEC laboratories in Leuven, Belgium. Marco's main interests and contributions are on compilers and programming models for a wide variety of processors and accelerators, processor architectures, multiprocessing and heterogeneous systems. An extended abstract of Marco's career has been published in HiPEAC Info #48, page 34.)