Tarek Abdelzaher (Office: SC 4126)
Office Hours: Wednesdays, 2pm-3pm, via zoom (unless otherwise requested).  

Zoom link: Join Zoom Meeting   

Tuesdays and Thursdays, 2:00 – 3:15pm (SC 0216)

In contrast to more general introductions to the Internet of Things (IoT), this advanced topic course focuses specifically on research challenges in edge AI motivated by the introduction of machine intelligence into IoT applications. Recent advances in AI culminate a shift in science and engineering away from strong reliance on algorithmic and symbolic knowledge towards new data-driven approaches. The course discusses how the emerging intelligent data-centric world impacts research on IoT and embedded computing. It organizes these effects around the types of bottlenecks that arise. At training time, in intelligent IoT applications, the bottlenecks are generally data related. IoT applications often exploit scarce data modalities (such as heterogeneous sensor data), unlike those commonly addressed in mainstream AI, necessitating solutions for efficient learning from scarce sensor data. At inference time, the bottlenecks are resource-related, calling for smaller models (such as small language models and application-specific foundation models) and improved resource economy (thanks to a variety of optimization techniques including quantization, caching, early exit networks, and mixture-of-expert scheduling policies). Furthermore, the convergence of AI around specific model architectures introduces additional model-related challenges in IoT contexts. The class discusses the research directions that arise in the data-centric world of intelligent IoT, covering data-, resource-, and model-related challenges, and overviews recent solutions emerging in this important domain. 

The course is designed for graduate students and grading reflects that design. Grades will be assigned as follows:

• - Class Project: 60% of the total grade is on a group project that produces a novel AI-based tool for IoT applications (groups of 3-4 students) 
• - Paper Presentations: 10% of the total grade. They are done in groups of 3-4 students.
• - Homework: 20% of the grade. 
• - Class Participation: Accounts for 10%. Includes active contributions to a positive class atmosphere and  class discussion.