Duration: 19:39

311 views

2

Accelerating software 2.0 for IoT & Edge

​

Researchers and scientists are pushing the boundaries of AI to create more complex models for image recognition, NLP, and other exciting use cases, but they are limited by current conventional system architectures. AI will be as transformative for the world as Henry Ford’s automobile was a hundred years ago, but to get there we need a new computer architecture that can run ever larger and more complex models, and that puts AI within reach of all organizations, not just the hyperscale companies that dominate the field today. In this session, Kunle will describe some of the exciting ways that researchers are pushing the limits of AI and what they need to go beyond the current constraints and allow AI to achieve its full potential.



With current state-of-the-art systems, developers are forced to do complicated cluster programming for multiple racks of systems and to manually program data parallelism and workload orchestration. This requires extreme specialization that puts AI out of reach for many organizations, and still lacks the computational power for today’s massive AI models. This talk will describe the need for a new software and hardware systems architecture that better supports the dataflow models used by today’s machine learning frameworks, and that can eliminate the deficiencies caused by the instruction sets that bottleneck conventional hardware today. Such a system, with a simpler programming model, would allow organizations of all sizes to run big data models with ease and simplicity.



Kunle will draw on his experience as founder of Afara Websystems (acquired by Sun Microsystems in 2002), as a Stanford University Electrical Engineering Professor, and as the current Chief Technologist of SambaNova Systems.

​

Kunle Olukotun, Chief Technologist & Co-Founder, SambaNova Systems and Cadence Design Professor of Electrical Engineering and Computer Science, Stanford University