Given several related tasks, multi-task learning (MTL) learns those tasks jointly by exploring the interdependence between them. Traditional multi-task learning methods mainly have two ways to measure task relatedness: sharing common parameters or sharing common features. However, both of them assume that all tasks are related and the strength of relatedness between tasks is the same. In real world, this is not often the case because of the complexity of the data. In this paper, we propose a graph-based multi-task learning method which measures the relatedness between tasks via a graph. The relatedness between tasks and the strength of the relatedness will be learned automatically. Experimental results demonstrate the effectiveness of our proposed graph-based multi-task learning method.
展开▼
