analiza aiqbench
DuoGNN addresses critical expressiveness bottlenecks in Graph Neural Networks by decoupling homophilic and heterophilic interactions to mitigate over-smoothing and over-squashing effects. This architecture enables scalable processing of large graphs while capturing both short-range and long-range dependencies, offering a robust solution for complex medical imaging tasks like disease diagnosis. The method demonstrates consistent performance improvements across diverse node classification datasets without incurring prohibitive computational costs.
DuoGNN: Topology-aware Graph Neural Network with Homophily and Heterophily Interaction-Decoupling
arXiv:2409.19616v3 Announce Type: replace Abstract: Graph Neural Networks (GNNs) have proven effective in various medical imaging applications, such as automated disease diagnosis. However, due to the local neighborhood aggregation paradigm in message passing which characterizes these models, they inherently suffer from two fundamental limitations: first, indistinguishable node embeddings due to heterophilic node aggregation (known as over-smoothing), and second, impaired message passing due to aggregation through graph bottlenecks (known as over-squashing). These challenges hinder the model expressiveness and prevent us from using deeper models to capture long-range node dependencies within the graph. Popular solutions in the literature are either too expensive to process large graphs due to high time complexity or do not generalize across all graph topologies. To address these limitations, we propose DuoGNN, a scalable and generalizable architecture which leverages topology to decouple homophilic and heterophilic edges and capture both short-range and long-range interactions. Our three core contributions introduce (i) a topological edge-filtering algorithm which extracts homophilic interactions and enables the model to generalize well for any graph topology, (ii) a heterophilic graph condensation technique which extracts heterophilic interactions and ensures scalability, and (iii) a dual homophilic and heterophilic aggregation pipeline which prevents over-smoothing and over-squashing during the message passing. We benchmark our model on medical and non-medical node classification datasets and compare it with its variants, showing consistent improvements across all tasks. Our DuoGNN code is available at https://github.com/basiralab/DuoGNN.
Abstract:Graph Neural Networks (GNNs) have proven effective in various medical imaging applications, such as automated disease diagnosis. However, due to the local neighborhood aggregation paradigm in message passing which characterizes these models, they inherently suffer from two fundamental limitations: first, indistinguishable node embeddings due to heterophilic node aggregation (known as over-smoothing), and second, impaired message passing due to aggregation through graph bottlenecks (known as over-squashing). These challenges hinder the model expressiveness and prevent us from using deeper models to capture long-range node dependencies within the graph. Popular solutions in the literature are either too expensive to process large graphs due to high time complexity or do not generalize across all graph topologies. To address these limitations, we propose DuoGNN, a scalable and generalizable architecture which leverages topology to decouple homophilic and heterophilic edges and capture both short-range and long-range interactions. Our three core contributions introduce (i) a topological edge-filtering algorithm which extracts homophilic interactions and enables the model to generalize well for any graph topology, (ii) a heterophilic graph condensation technique which extracts heterophilic interactions and ensures scalability, and (iii) a dual homophilic and heterophilic aggregation pipeline which prevents over-smoothing and over-squashing during the message passing. We benchmark our model on medical and non-medical node classification datasets and compare it with its variants, showing consistent improvements across all tasks. Our DuoGNN code is available at this https URL.
Machine Learning (cs.LG); Social and Information Networks (cs.SI)
arXiv:2409.19616 [cs.LG]
(or arXiv:2409.19616v3 [cs.LG] for this version)
Mancini, K., Rekik, I. (2025). DuoGNN: Topology-Aware Graph Neural Network with Homophily and Heterophily Interaction-Decoupling. Graphs in Biomedical Image Analysis. GRAIL 2024, vol 15182. Springer
https://doi.org/10.1007/978-3-031-83243-7_12 DOI(s) linking to related resources
Text integral sub licența deschisă a publicației · 275 cuvinte
Subiecte conexe
Subgraph Filtering for Fair Graph Neural Networks
arXiv · 28 aug. 2026
Not All Nodes Are Created Equal: Homophily-Aware Stratification for Stable GNN Evaluation
arXiv · 18 sept. 2026
Topology-Aware Gaussian Graph Repair for Robust Graph Neural Networks
arXiv · 3 iun. 2026
Towards Near-Real-Time Telemetry-Aware Routing with Neural Routing Algorithms
arXiv · 9 sept. 2026