Pratik Jawanpuria, Maksim Lapin, Matthias Hein, Bernt Schiele
TL;DR该研究通过正半定核理论，提出一种可以优化多任务学习的无约束对偶问题，从而节省了计算时间。
Abstract
The paradigm of multi-task learning is that one can achieve better generalization by learning tasks jointly and thus exploiting the similarity between the tasks rather than learning them independently of each other. While previously the relationship between tasks had to be user-defined