On-device Training and Transfer Learning - Lecture 21

Explore on-device training and transfer learning concepts in this MIT lecture delivered by Prof. Song Han, focusing on critical aspects of efficient machine learning implementation. Delve into the security concerns of gradient sharing through deep leakage analysis, and understand how to overcome memory constraints in on-device training scenarios. Learn about innovative approaches including Tiny Transfer Learning (TinyTL), Sparse Back-propagation (SparseBP), and Quantized Training with Quantization Aware Scaling (QAS). Discover the implementation of PockEngine and its system support for sparse back-propagation, gaining practical insights into making machine learning more efficient and secure for on-device applications.