Multi-task Graph Neural Architecture Search with Task-aware Collaboration and Curriculum

Graph neural architecture search (GraphNAS) has shown great potential for automatically designing graph neural architectures for graph related tasks. However, multi-task GraphNAS, capable of handling multiple tasks simultaneously and capturing the complex relationships and dependencies between them, has been largely unexplored in literature. To tackle this problem, we propose a novel multi-task graph neural architecture search with task-aware collaboration and curriculum (MTGC 3 ), which is able to simultaneously discover optimal architectures for different tasks and learn the collaborative relationships among different tasks in a joint manner. Specifically, we design the structurally diverse supernet to manage multiple architectures and graph structures in a unified framework, which combines with our proposed soft task-collaborative module to learn the transferability relationships between tasks. To further improve the architecture search procedure, we develop the task-wise curriculum training strategy that reweighs the influence of different tasks based on their relative difficulties. Extensive experiments show that our proposed MTGC 3 model achieves state-of-the-art performance against several baselines in multi-task scenarios, demonstrating its ability to discover effective architectures and capture collaborative relationships for multiple tasks.