AUTHOR=Vijayan Priyesh , Chandak Yash , Khapra Mitesh M. , Parthasarathy Srinivasan , Ravindran Balaraman 

TITLE=Scaling Graph Propagation Kernels for Predictive Learning

JOURNAL=Frontiers in Big Data

VOLUME=Volume 5 - 2022

YEAR=2022

URL=https://www.frontiersin.org/journals/big-data/articles/10.3389/fdata.2022.616617

DOI=10.3389/fdata.2022.616617

ISSN=2624-909X

ABSTRACT=Many real-world applications deal with data that have an underlying graph structure associated with them. Efficiently capturing the relation of a node and its expanded neighborhood is thus crucial to perform any downstream analysis on this data. Specifically, we look at the problem of Collective Classification (CC) for assigning labels to unlabeled nodes. Most of the state-of-the-art models for CC make extensive use of differentiable variants of Weisfeiler-Lehman (WL) kernels. Due to limitations of current computing architectures, WL kernels are limited in their ability to capture useful relations over the expanded neighborhood of a node. To address this concern, we propose a framework, I-HOP, that couples iterative inference mechanism with differentiable kernels. It iteratively levers the summarized label information in conjunction with the attribute information to consider much larger hops. Additionally, we point out a limitation of WL kernels where the node's original information is decayed exponentially with an increase in neighborhood size, and provide a solution to address it. Extensive evaluation across 11 datasets showcases the improved results and robustness of our proposed iterative framework, I-HOP.