Abstract: Feature selection can eliminate noise and redundant information in data, simplify computational complexity and data analysis difficulty, so it has significant research value in data mining and machine learning. With the development of deep learning technology, deep neural networks have been applied to feature selection and achieved better results than traditional methods. Still, there is a lack of comprehensive description and discussion of such research. In this paper, we described the traditional feature selection algorithms, and summarized the research progress of deep feature selection algorithms into two categories: input-layer embedding and encoding-layer embedding. The effects of typical algorithms were tested on public datasets, and challenging and research directions were further discussed.