Adaptive Self-Supervised Graph Representation Learning

Graph neural networks (GNNs) is widely used to learn a powerful representation of graph-structured data. Recently, many works shows that transferring knowledge from self-supervised tasks to downstream tasks can improve graph representation. However, in terms of optimization target and training data, there is a fundamental difference between self-supervised and downstream tasks. The reason is that these methods do not make any adaptations for downstream activities, traditional pre-training approaches may not be successful enough for knowledge transfer. To address these issues, we present a generic learning framework that identifies the solution in graph self-supervised learning where the primary task, rather than auxiliary tasks, obtains the highest performance. Extensive experiments on a variety of datasets show that the proposed framework may efficiently integrate auxiliary tasks with the main task and enhance performance when compared to state-of-the-art methods.