Watch a 59-minute seminar from the Federated Learning One World series featuring MIT's Song Han as he explores the challenges and solutions for on-device training with extremely limited memory constraints of 256KB. Learn about innovative approaches to machine learning deployment on resource-constrained devices, drawing from Han's expertise in efficient AI systems and tiny machine learning. Discover practical techniques for implementing neural network training in environments where memory is at a premium, making AI accessible on edge devices and microcontrollers.