Save Big on Coursera Plus. 7,000+ courses at $160 off. Limited Time Only! Grab it Learn advanced distributed training concepts in this MIT lecture covering hybrid parallelism, auto-parallelization techniques, and strategies for overcoming bandwidth and latency bottlenecks in machine learning systems. Explore gradient compression methods including gradient pruning with sparse communication and deep gradient compression, as well as gradient quantization approaches like 1-Bit SGD and TernGrad. Understand how delayed gradient updates can address latency challenges in distributed training environments. Delivered by Professor Song Han as part of the MIT 6.5940 course, this 59-minute lecture provides essential knowledge for implementing efficient distributed machine learning systems.