Search For Deep Graph Neural Networks
This work addresses the challenge of building deep GNNs for graph-based tasks, offering an incremental improvement over current GNN-oriented NAS methods by enabling deeper architectures.
The paper tackles the problem of designing deep graph neural networks (GNNs) by proposing a two-stage search space and generation pipeline to automatically create high-performance, transferable models, while addressing the 'over-smooth' issue with residual connections and identity mapping; experiments show that the generated models outperform existing manually designed and NAS-based ones on real-world datasets.
Current GNN-oriented NAS methods focus on the search for different layer aggregate components with shallow and simple architectures, which are limited by the 'over-smooth' problem. To further explore the benefits from structural diversity and depth of GNN architectures, we propose a GNN generation pipeline with a novel two-stage search space, which aims at automatically generating high-performance while transferable deep GNN models in a block-wise manner. Meanwhile, to alleviate the 'over-smooth' problem, we incorporate multiple flexible residual connection in our search space and apply identity mapping in the basic GNN layers. For the search algorithm, we use deep-q-learning with epsilon-greedy exploration strategy and reward reshaping. Extensive experiments on real-world datasets show that our generated GNN models outperforms existing manually designed and NAS-based ones.