Learning multi-task communication with message passing
Pengfei Liu, Jie Fu*, Yue Dong*, Xipeng Qiu, Jackie Chi Kit Cheung
Fig. Different topology structures for multi-task learning: flat, hierarchical and graph structures. Each blue circle
represents a task (A, B, C, D), while the red box denotes a virtual node, which stores shared information and facilitates communication
Contributions
We explore the problem of learning the relationship between multiple tasks and formulate this problem as message passing over a graph neural network.
We borrow ideas from interaction systems and propose new architectures for multi-task learning Complete-graphs and Star-graphs .
Different from traditional black-box learned models, this paper makes a step towards learning transferable and interpretable representations, which enables us to know what types of patterns are shared
Case Study
Fig. : Illustrations of the interpretable patterns captured by the
shared layer under different tasks.