StepGCN: Step-oriented Graph Convolutional Networks in Representation Learning

Juyeob Lee, Minyoung Lee, Eunil Park

Abstract: Graph Convolutional Networks (GCNs) are employed to address a number of tasks in our society with their representation learning approach. Nonetheless, despite their effectiveness and usefulness, the majority of GCN-oriented approaches have an over-smoothing concern. Over-smoothing is the problem of node representations converging into a certain value, making the nodes indistinguishable. To effectively address the over-smoothing problem, we introduce StepGCN, a GCN model that integrates step learning techniques with graph residual connection networks. With our StepGCN, we achieved significant performance improvements in multiple representation learning benchmark datasets, and demonstrate that step learning can be expanded to other graph networks.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: Deep Learning and representational learning