Jiayi Shen

Multi-task learning is a fundamental learning paradigm for machine learning, which aims to simultaneously solve multiple related tasks to improve the performance of individual tasks by sharing knowledge. The crux of multi-task learning is to explore task relatedness to improve each individual task, which is non-trivial since the underlying relationships among tasks can be complicated and highly nonlinear. Early works deal with this crux by learning shared features, designing regularizers imposed on parameters or exploring priors over parameters. Recently, multi-task deep neural networks have been developed, learning shared representations in the feature layers while keeping classifier layers independent. Recently, multi-task learning has generated increasing interest in medical image analysis, e.g., multiple organ localization and direct volume estimation. In the real world, the great power of multi-task learning in medical image analysis has largely been under-developed, due to the data insufficient problem. Therefore, when very limited training data is given, it is highly desired to propose new general frameworks for multi-task learning. In this case, it is difficult to learn a proper model for each task independently without overfitting.