Algorithm-System Co-design for Efficient and Hardware-Aware Embedded Machine Learning

We introduce efficient algorithm and system co-design for embedded machine learning to reduce the memory and computation cost. The software stack includes two sides: the inference system and the deep learning algorithms. We first introduce current inference libraries and scheduling optimization techniques. Then, we introduce efficient deep learning model designs, including model compression and neural architecture search. Finally, we discuss how we can co-design the system and algorithm to achieve the best end-to-end performance.