On-device Training and Transfer Learning in Machine Learning - Lecture 21

Learn about on-device training and transfer learning in this recorded MIT lecture that explores critical aspects of machine learning deployment. Dive into the challenges and solutions surrounding gradient security, memory constraints in on-device training, and advanced techniques like Tiny Transfer Learning (TinyTL) and Sparse Back-propagation (SparseBP). Master the implementation of quantized training using Quantization Aware Scaling (QAS), understand the risks of deep leakage from gradients, and explore system support for sparse back-propagation through PockEngine. Professor Song Han guides you through these essential concepts that are fundamental to developing efficient and secure on-device machine learning systems.