Multi-task learning (MTL) has proven successful in various applications of machine learning, including natural language processing, speech recognition, computer vision, and drug discovery. This article provides an overview of MTL, particularly in deep neural networks, and discusses recent advances. It introduces two common methods for MTL in deep learning—hard and soft parameter sharing—and explains why MTL works in practice. The article also reviews the literature on MTL and recent methods for MTL in deep neural networks, such as Deep Relationship Networks, Fully-Adaptive Feature Sharing, and Cross-stitch Networks. Additionally, it explores the use of auxiliary tasks and the challenges of task similarity and relationship in MTL. The article concludes by highlighting the need for a more principled understanding of task similarity to improve the generalization capabilities of MTL in deep neural networks.Multi-task learning (MTL) has proven successful in various applications of machine learning, including natural language processing, speech recognition, computer vision, and drug discovery. This article provides an overview of MTL, particularly in deep neural networks, and discusses recent advances. It introduces two common methods for MTL in deep learning—hard and soft parameter sharing—and explains why MTL works in practice. The article also reviews the literature on MTL and recent methods for MTL in deep neural networks, such as Deep Relationship Networks, Fully-Adaptive Feature Sharing, and Cross-stitch Networks. Additionally, it explores the use of auxiliary tasks and the challenges of task similarity and relationship in MTL. The article concludes by highlighting the need for a more principled understanding of task similarity to improve the generalization capabilities of MTL in deep neural networks.