Deep Multi-task Augmented Feature Learning via Hierarchical Graph Neural Network

Deep multi-task learning attracts much attention in recent years as it achieves good performance many applications. Feature is important to deep for sharing common information among tasks. In this paper, we propose a Hierarchical Graph Neural Network (HGNN) learn augmented features learning. The HGNN consists of two-level graph neural networks. the low level, an intra-task network responsible powerful representation each data point task by aggregating its neighbors. Based on learned representation, embedding can be generated similar way max pooling. second inter-task updates embeddings all tasks based mechanism model relations. Then one used augment feature points task. Moreover, classification tasks, inter-class introduced conduct operations finer granularity, i.e., class generate using representation. proposed augmentation strategy models. Experiments real-world datasets show significant improvement when strategy.