We are delighted to present a new virtual special issue of Stat co-edited by Hao Helen Zhang and Yufeng Liu entitled “Deep Learning from Statistical Perspectives,” which aims at gathering literature reviews, recent advances, and novel contributions in deep learning, from various statistical perspectives, to inspire new ideas and promote state-of-the-art research. The 11 high-quality articles cover a broad range of topics on theory, computation, implementation, and applications of deep learning. They are compiled as an overview and three sections:

(1) Theory of deep learning;

(2) Improvements of DNNs on computation and implementation: training, tuning, and variable selection; and

(3) Connections between deep learning and classical statistical models.

Deep learning is a fast-growing subfield of machine learning. It has gained high popularity and received a lot of attention in the media and scientific community. In particular, deep learning has been successfully used in a large variety of applications, ranging from image recognition, natural language processing, to bioinformatics and board games. It has shown exceptional performance and great flexibility in applications.