Robust Temporal Smoothness in Multi-Task Learning

Multi-task learning models based on temporal smoothness assumption, in which each time point of a sequence points concerns task prediction, assume the adjacent tasks are similar to other. However, effect outliers is not taken into account. In this paper, we show that even only one outlier will destroy performance entire model. To solve problem, propose two Robust Temporal Smoothness (RoTS) frameworks. Compared with existing relation, our methods chase information but identify tasks, however, without increasing computational complexity. Detailed theoretical analyses presented evaluate methods. Experimental results synthetic and real-life datasets demonstrate effectiveness We also discuss several potential specific applications extensions RoTS