Wei Li - The World’s Deep Learning Runs on CPUs
The World’s Deep Learning Runs on CPUs
The conventional wisdom is that deep learning (DL) has to be done on accelerators. However, the world’s deep learning runs very well on general purpose CPUs today, adaptable to deep learning that continues to evolve quickly from its early success in CNNs, to the adoption of RNNs, Recommendation Systems, and Reinforcement Learning with a wide range of performance characteristics for computer systems in production environments. In this talk, I will cover the hardware and software technologies that power AI in the world’s clouds and data centers. I will show a 200X performance improvement, and how to reduce inference to milliseconds and scale training to 1000s of compute nodes.
Host: Andrew Chien
Wei Li
Wei Li is Vice President and General Manager of Machine Learning and
Translation at Intel Corporation. In his 20 years at Intel, he has led
innovative software efforts for artificial intelligence, clouds/data centers,
mobile devices, and hardware/software co-design. He served as an
associate editor for ACM Transactions on Programming Languages and
Systems, and was on the joint Intel/Microsoft technical committee that
funded the creation of parallel computing research centers at the University
of California-Berkeley and the University of Illinois at Urbana-Champaign.
He received a Ph.D. in Computer Science from Cornell University, and
completed an executive program at Stanford Graduate School of Business.
