This chapter presents the research topics in serverless machine learning. The model training procedure of Machine Learning (ML) usually consumes large amounts of time and cost, and need to be improved. Following the footsteps of conventional ML architecture, the emerging serverless computing paradigm brings new opportunities for ML to mitigate running time and operational overheads. Moreover, the ML developers then only need to define and manage the stateless functions, rather than the dedicated servers or middleboxes. Hence, serverless machine learning is envisioned by global researchers as a promising trend of ML techniques and applications.