Overview
This course teaches learners how to implement model merging for memory-efficient, real-time video analytics at the edge. The main goals of the course are to reduce memory usage and improve accuracy in edge deployments of video analytics pipelines. The course covers the skills of identifying merging configurations, altering inference schedules, and integrating merging into existing pipelines. The teaching method involves presenting a new memory management technique called model merging and explaining its benefits through experiments. This course is intended for individuals interested in video analytics, edge computing, and memory optimization techniques for real-time inference.
Syllabus
NSDI '23 - Gemel: Model Merging for Memory-Efficient, Real-Time Video Analytics at the Edge
Taught by
USENIX