Data Science Institute | Sense, Collect & Move Data Center Seminar
Speaker: Mingoo Seok, Columbia University
The interest in the Internet of Things (IoT) and the cognitive computing via machine learning is rapidly rising, and it is very natural for many to envision the combination of those two. In several applications, such combination promises to enable new features, to improve accuracy in digital processing and autonomous decision, and to reduce wireless communication bandwidth and thus system power dissipation. Among several candidates, neural network (NN) based systems gain significant attention for several desirable characteristics including high accuracy, regularity, parallelism, and programmability. However, it indeed poses several challenges to include/implement such cognitive functions in/for IoT sensing devices due to the limited resources (hardware and energy) available in those devices. In this seminar, we will discuss those challenges, namely implementing machine-learning in resource-constrained IoT devices, and present our recent efforts across algorithm, hardware architecture, and circuits, and some combinations of those, to address the challenge.