Deep Learning, a core of modern Artificial Intelligence, is rapidly expanding to resourceconstrained devices, including smartphones, wearables, and intelligent embedded systems for improving response time, privacy, and reliability. This course focuses on bringing these powerful deep-learning applications from central data centers and large GPUs to distributed ubiquitous systems. On-Device Deep Learning is an interdisciplinary topic at the intersection of artificial intelligence and ubiquitous systems, dedicated to enabling computing on edge devices. This course includes a wide range of topics related to deep learning in resource constrained settings including pruning and sparsity, quantization, neural architecture search, knowledge distillation, on-device training and transfer learning, distributed training, gradient compression, federated learning, efficient data movement and accelerator design, dynamic network inference, and advanced compression and approximation techniques for enabling on-device deep neural network inference and training. This course provides a comprehensive foundation for cutting-edge “tinyML” expertise
The students should have an introductory undergraduate-level or graduate-level introductory background in machine learning and deep neural networks.