Federated Learning Application on Depression Treatment Robots(DTbot)

Abstract

Depression is one of the most prevalent psychiatric disorders and an important public health problem. Its etiology is multifaceted, and the specific pathophysiological mechanisms are still unclear. At present, the main treatment methods for depression are medication, psychotherapy and physical therapy, and clinical applications usually combine two or three methods. Psychotherapy is currently mainly oriented towards the traditional face-to-face communication with psychologists, and is rarely combined with the current rapid development of technology. In this paper, we aim to design an intelligent robot that incorporates deep learning methods to help doctors treat patients more efficiently. The problem is that the current models of robots are trained by uploading data to a server, and then having the server train the robot. There are disadvantages of this approach. First, patient videos and conversations are private information. So uploading those private information to the server can lead to patient information leakage, which is bad. Second, the data recorded in daily life, including audio and video, are very large files that are slow to transfer and tend to cause package loss and other problems in the process. Training a multi-robot model in combination with federal learning would be a good solution to these two problems. The article combines federal learning with basic deep learning methods to design a depression treatment robot(DTbot) that can treat patients with more privacy and efficiency while handling their personal information.