Organizers
- Alessandro Sperduti, University of Padova, Italy
- Benoit Gaüzère, INSA Rouen Normandie
- Caterina Graziani, University of Siena
- Davide Rigoni, University of Padova
- Domenico Tortorella, University of Pisa
- Filippo Maria Bianchi, UiT the Arctic University of Norway
- Matteo Tolloso, University of Pisa
- Sara, Bacconi, University of Siena
- Vincenzo Carletti, University of Salerno
Abstract
Graphs play a crucial role in different fields in modeling complex structures composed of entities and their relationships, including dynamic domains where these relationships can evolve over time. Notable examples of graph-based representations and processing can be found in biology, where the structures of molecules and proteins are naturally modeled as graphs; social sciences, where graphs are used to model interactions between individuals or groups; data science, where graphs enhance recommendation systems by tracking user-item interactions; and transportation, where graphs are employed to model the evolution of traffic flow over time.
Neural models on graphs enable adaptive solutions for a wide range of learning tasks on graph data, avoiding the need for hand-engineered features or domain-specific knowledge. This capability has driven significant progress in applying machine learning to graph-based problems across various research fields. As a result, the design, optimization, and analysis of these graph-based learning models have become central to cutting-edge research, while also presenting a range of open challenges that continue to shape the field’s future directions.
This special session at ICANN 2026 aims to bring together cutting-edge research and new ideas in neural networks and machine learning models for graphs. We encourage the submission of works that address open challenges by advancing both theoretical investigations and practical applications.
List of Topics Covered in the Special Session
- Graph neural networks based on convolutional, recurrent, and transformer architectures
- Temporal and dynamic graphs
- Relational inference, heterogeneous graphs
- Graph pooling, graph structure learning
- Open problems in representation learning, e.g. over-smoothing, over-squashing, heterophily
- Graph learning for time series, including data imputation
- Theory of graph learning
- Graph signal processing, including spectral methods for analysis and design
- Trustworthy AI for graph learning, including explainability (XAI), robustness, reliability
- Graph-based methodologies for pattern recognition
- Other methods for learning on graphs, including kernel-based approaches
- Datasets and benchmarks for learning on graphs
- Applications of graph learning, including: Chemistry and biology, e.g. toxicology, protein interactions; Graph learning on brain data; Social sciences, social networks; Graph learning for ecology; Knowledge engineering and discovery; Sensor networks and IoT applications