Lecture, four hours; discussion, two hours; outside study six hours. Requisite: course M146. To alleviate costs and improve robustness and generalization performance of modern machine learning models, it becomes crucial to develop methods with strong theoretical guarantees to warrant efficient, accurate, and robust learning. Discussion of advanced topics and state-of-art research to improve efficiency, robustness, and scalability of machine learning algorithms on large data. Topics include advanced optimization, variance reduction, distributed training, federated learning, data summarization, robust learning, neural network pruning, neural architecture search, neural network quantization. Letter grading.