During the past decade, Deep Learning based AI technology not only becomes the predominant solutions for existing or new problems, also almost instantly deployed to various smart devices. In this talk, we start with a brief review on how power-efficient AI engine helped this new AI wave and effectively enabled billions of battery-powered devices; then, we touch the new trend: always-on or long-continuous-run AI use cases, which require optimal minimum power solution. We discuss some details of ultra-low-power AI solution and how it offers the improved quality for targeted use cases. With continuous evolution of new intelligent algorithms, this talk concludes with on-going challenges and some potential directions.
Schedule
Timezone: PST
Evolutionary Needs of TinyML
Liang SHEN, Sr. Director of Engineering
Qualcomm
During the past decade, Deep Learning based AI technology not only becomes the predominant solutions for existing or new problems, also almost instantly deployed to various smart devices. In this talk, we start with a brief review on how power-efficient AI engine helped this new AI wave and effectively enabled billions of battery-powered devices; then, we touch the new trend: always-on or long-continuous-run AI use cases, which require optimal minimum power solution. We discuss some details of ultra-low-power AI solution and how it offers the improved quality for targeted use cases. With continuous evolution of new intelligent algorithms, this talk concludes with on-going challenges and some potential directions.
Liang SHEN, Sr. Director of Engineering
Qualcomm
Liang Shen joined Qualcomm from AMD/ATI as part of acquisition in 2009 and has been leading the development of multimedia software components for Snapdragon products. He and his team have been focusing on AI Processors and System since 2016. Liang got his B.Sc. on Biomedical Engineering and M.Sc./Ph.D. on Image Processing and Pattern Recognition with over 20 publications in journals and conferences. He worked on real-time signal & data processing algorithms and systems for radar, sonar, and satellites. He then led the development and successful commercialization of new-generation communication systems with TTS and ASR. In ATI/AMD, Liang was responsible for next-gen ASIC software and handheld software. While interested and enjoyed with all the projects/products he involved/developed, the most unimaginable one is cycling back to work on AI, his fantasy area – allowing him pursuing dream to make IoT becoming Wisdom-of-Things (WoT).
TinyML can be used to enrich courses across the STEM curriculum, ranging from machine learning to embedded systems, with exciting, hands-on projects. tinyMLedu is working to help widen access to such applied machine learning experiences by building an international coalition of researchers and practitioners. Through collaborations across academia and industry, we are working to develop and share high quality, open-access educational materials globally and provide global access to the requisite hardware and software resources. You can learn more about our efforts at tinyMLedu.org.
Schedule subject to change without notice.