Enhancing Subgraph-GNNs by means of Edge-Diploma Ego-Group Encodings
Authors: Nurudin Alvarez-Gonzalez, Andreas Kaltenbrunner, Vicenç Gómez
Abstract: We present a novel edge-level ego-network encoding for finding out on graphs which will improve Message Passing Graph Neural Networks (MP-GNNs) by providing additional node and edge choices or extending message-passing codecs. The proposed encoding is sufficient to differentiate Strongly Frequent Graphs, a family of adverse 3-WL equal graphs. We current theoretically that such encoding is additional expressive than node-based sub-graph MP-GNNs. In an empirical evaluation on 4 benchmarks with 10 graph datasets, our outcomes match or improve earlier baselines on expressivity, graph classification, graph regression, and proximity duties — whereas lowering memory utilization by 18.1x in positive real-world settings
