Title: Challenges and Progress on Advancing, Trusting, and Scaling AI
Artificial Intelligence (AI) is hot again thanks in part to recent advances in artificial neural networks and deep learning. As a result, the AI field is achieving unprecedented performance on long-standing tasks in computer vision, natural language processing, language translation, speech transcription and synthesis, conversational systems, and more. While further advancements on core AI methods will continue, important challenges remain for ensuring the development of trustworthy AI in practice and enabling a broader scope of AI deployments. Towards these goals, we present ongoing efforts at IBM Research on advancing AI that address fundamental challenges in learning more from less, mastering language, and neuro-symbolic reasoning. We describe IBM’s work on trusting AI, which includes fairness, explainability, robustness, and transparency. Finally, we present IBM’s efforts for scaling AI, which includes automation of AI model development and AI lifecycle management.
Bio of the speaker
Dr. John R. Smith is an IBM Fellow and Head of AI Tech at IBM T. J. Watson Research Center. He manages research at IBM T. J. Watson Research Center on computer vision, speech, language, knowledge and interaction and is IBM’s global lead on vision and video.
Dr. Smith has published several hundred articles (h-index = 74, i-10 index = 400) at top conferences and journals on computer vision, multimedia, and databases. Historically, with Prof. Shih-Fu Chang at Columbia University, Dr. Smith developed some of the earliest approaches for content-based image and video retrieval, including the VisualSEEk content-based image retrieval system, and WebSEEk, which was one of the first image and video search engines for the Web in 1995!
Dr. Smith was Editor-in-Chief of IEEE Multimedia from 2010 to 2014 and is a Fellow of IEEE.
Title: Enabling the SmartGrid with IoT Sensors and Edge-Cloud Analytics
Wireless sensors and edge-cloud analytics have the potential to gather and process vast amounts of data about the physical world, offering radical new insights about everything from critical infrastructure to interpersonal interactions. But designing, deploying, and operating geographically-distributed systems consisting of a hierarchy of sensing, storage, compute, and communication elements raises interesting new challenges across the system stack. In this talk, we will discuss our experiences designing IoT systems to address several power and power grid monitoring problems. In particular, this talk will focus on three systems-—PowerBlade, Triumvi, and GridWatch—-and their motivation, design, and deployment. PowerBlade explores how to cost-effectively characterize, capture, and classify widespread plug-load energy usage-—representing the fastest growing and least understood segment of end-use energy consumption-—across hundreds of homes and offices representing tens of thousands of sensors. Triumvi explores how to make circuit level energy metering, useful for a variety of facilities trending, energy savings, and fault detection & diagnostics applications, more efficient and scalable. Finally, GridWatch explores how to scalably and cost-effectively detect and respond to the power outages that stymie residential and business activity in under-developed power grids using mobile and fixed sensors, data analytics, and reporting systems in Sub-Saharan Africa, finding that conventional approaches to outage detection vastly underreport customer experiences. The systems all share similar architectures, require new sensor devices and edge-cloud data processing, and wrestle with power management and networking. But they ultimately demonstrate both the tremendous potential and significant challenges of this nascent computing class.
Bio of the speaker
Prabal Dutta is an Associate Professor of Electrical Engineering and Computer Sciences at the University of Calfornia at Berkeley, where he co-directs the CONIX Research Center. Previously, he was a Morris Wellman Faculty Development Associate Professor of Electrical Engineering and Computer Science at the University of Michigan. His interests span circuits, systems, and software, with a focus on mobile, wireless, embedded, networked, and sensing systems with applications to health, energy, and the environment. His work has yielded dozens of hardware and software systems, has won five Top Pick/Best Paper Awards, two Best Paper Nominations, and a Potential Test of Time 2025 Award, as well as several demo, design, and industry competitions. His work has been directly commercialized by a dozen companies and indirectly by many dozens more, has been utilized by thousands of researchers and practitioners worldwide, and is on display at Silicon Valley’s Computer History Museum. His research has been recognized with an Sloan Fellowship, a CAREER Award, a Popular Science Brilliant Ten Award, an Intel Early Career Faculty Fellowship, and as a Microsoft Research Faculty Fellowship Finalist. He has served as chair or co-chair of MobiSys’18, BuildSys’17, IPSN’17, ESWEEK’17 IoT Day, HotMobile’16, SenSys’14, and HotPower’11, and on the DARPA ISAT Study Group from 2012-2016, where he co-chaired numerous studies. He holds a Ph.D. in Computer Science from UC Berkeley (2009), where NSF and Microsoft Research Graduate Fellowships supported his research. He received an M.S. in Electrical Engineering (2004) and a B.S. Electrical and Computer Engineering (1997), both from The Ohio State University.