Abstract: Digital predistortion (DPD) techniques are now widely used to correct the nonlinearity of power amplifiers (PAs) and reduce power dissipation in the transmitter front-ends. With the development of communication technology, high-performance and low-complexity DPD technology has become a hot spot of current research. The development of machine learning (ML) provides new ideas for research and plays an important role in the development process of DPD. Based on ML, this paper focuses on the three research directions of model construction, parameter extraction and varying transmission configurations DPD, summarises the relevant literature, and elaborates the existing methods in each direction.