Scalable Machine Learning

Time: W 9-10:30am
Location: 320 Soda, 1-2 units
Instructor: John Canny

Class Piazza is open Links to an external site.

This course will cover systems and AI topics related to machine learning at very large scale. The course will cover deep learning and reinforcement learning as well as general machine learning models. We are particularly interested in deep learning and reinforcement learning on massive emerging datasets like camera data from computer assisted vehicles, drones etc. Single-machine training times on million-image datasets (ImageNet) are in the days. Effective parallelization methods have reduced that time to hours. But a single vehicle generates more than a million images each day, and we will soon be dealing with datasets millions or billions of times larger. This will require radically new approaches to training. Careful selection of training examples will be essential. "Learning to learn" inspired by human meta-cognition is a potential guiding principle.

The course will have one 1.5-hour meeting per week (Weds), and is available for one or two units of credit. For two unit credit, students will be expected to do a class project and presentation. First meeting Weds 1/20.