Problem
Research questions and friction points this paper is trying to address.
Graph Edit Distance
Generalization
Cross-collection Transfer
Neural Models
Revisiting the Generalization of Neural Graph Edit Distance Models
🤖 AI Summary
This study addresses the limited cross-dataset generalization of existing neural graph edit distance (GED) models, a performance gap obscured by conventional evaluation paradigms. To tackle this issue, the authors first quantify the cross-set transfer gap and propose a multi-source joint pretraining strategy leveraging exact GED supervision. Furthermore, they establish a comprehensive evaluation framework for zero-shot and few-shot transfer scenarios. The findings reveal significant performance degradation caused by discrepancies between training and test distributions. Crucially, the results demonstrate that multi-source pretraining substantially enhances target-domain performance and adaptability to novel tasks. By exposing these previously overlooked generalization limitations and validating an effective mitigation strategy, this work introduces a new paradigm for rigorously evaluating the generalizability of neural GED models across diverse domains.
📝 Abstract
Neural approaches to Graph Edit Distance (GED) have achieved strong results under standard within-dataset evaluation, but much less is known about how well these models transfer across graph collections. We conduct a systematic study of this problem using exact GED supervision across diverse graph datasets and a broad set of representative learning-based methods. Our results reveal a pronounced gap between within-collection performance and cross-collection transfer. Models that perform well on their training collections often lose this advantage when evaluated on structurally different data. Training on multiple source collections substantially improves zero-shot transfer and provides a better starting point when limited supervision is available for a new target collection. Further analysis shows that transfer behavior varies with the source--target direction and the structural characteristics of the collections involved. These findings suggest that conventional within-collection evaluation provides only a partial view of the generalization behavior of neural GED models and motivate broader evaluation across heterogeneous graph collections.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Graph Edit Distance
Cross-collection Generalization
Zero-shot Transfer
Neural GED Models
Heterogeneous Graphs
🔎 Similar Papers
No similar papers found.
