Recently, graph structures have attracted a lot of attention in many domains. Indeed, there is an increasing number of applications where data are cogently represented by well-structured and flexible graph models, mainly due to their ability to encode both topological and semantic information, thus going beyond the classical and simple Euclidean domain. For example, in e-commerce, a graph-based learning system can exploit the interactions between users and products to make highly accurate and customized recommendations. In chemistry, molecules are modelled as graphs, and identifying their bioactivity leads to drug discovery. In a citation network, papers are linked to each other via citationships, and they need to be categorized into groups. Beyond these examples, data can be represented by graphs in many other applications: scene graph generation and understanding, object tracking, point clouds classification, and action recognition in computer vision; proteinomic and genomic data, text classification, relationships among documents or words for inferring document labels in
natural language processing; forecasting traffic speed, congestion, and anomalies in transportation networks; and many more. Remarkably, in most of the above scenarios, the adoption of Graph Neural Network (GNN) models has been proven to be particularly effective.
Despite their success and wide applicability, some important issues, such as scalability, computing adaptability, and effectiveness of the solutions, still remain open:
Scalability – It is the main bottleneck in using graphs for real applications; graphs have been used from the beginnings in computer science domains such as operational research, pattern recognition, programming and computing, but most of the proposed methods and algorithms could handle sparse graphs only. More recently, many diverse solutions have been proposed to address this problem, such as heuristic methods, approximate solutions, and the exploitation of massively parallel and distributed architectures. Nonetheless, scalability issues still remain an open problem.
Computing adaptability – Recently, in many research domains such as pattern recognition, databases, and knowledge reasoning, we are witnessing a significant shift from Euclidean space representations towards graph-based ones. Nonetheless, this re-design poses problems of adaptability that deserve to be tackled. Moreover, there exist many other aspects to take into account when dealing with graphs, such as the memory utilization (e.g., wasting allocated memory with very large sparse matrices representing graphs), or the lack of locality principle in memory accesses, both limiting the performances of most graph algorithms.
Effectiveness of the solutions – In some domains, algorithms dealing with vector data are still far better performing than their graph-based version, in terms of accuracy, error rate, and other metrics. This poses the question whether such inefficiency is due to the graph-based representation itself or to the inadequacy of the algorithmic model designed to handle such data structures. To this end, novel algorithms and models adopting Deep Learning on Graphs may represent a promising direction requiring further investigation.
The purpose of this special section is to provide a forum for all novel aspects of graph-based methods over wide application and research domains, as well as to foster a thorough discussion about state-of-the-art techniques and results, achieved goals, and open challenges. Topics of interest to this special section include:
- Advances in computing and learning on graphs
- Graph Neural Networks (GNNs)
- Graph data fusion methods and graph embedding techniques
- Efficient, parallel, and distributed processing frameworks for big graphs
- Novel dynamic, spatial, and temporal graphs for recognition and learning
- Emerging graph-based methods in computer vision
- Interactivity, explainability, and trust in graph-based learning methods
- Applications of GNNs
- Human behavior and scene understanding using graphs
- Benchmarks for GNNs
- Graph signal processing
- Application of graph data processing in biology, healthcare, transportation, natural language processing, social networks, etc.
– deadline for submissions: 30 June 2022
– first decision (accept/reject/revise): 30 November 2022
– submission of revised papers: 15 February 2023
– notification of final decision: 30 April 2023
– journal publication: first half of 2023
For author information and guidelines on submission criteria, please visit the TETC Author Information page. Please submit papers through the ScholarOne system, and be sure to select the special-section name. Manuscripts should not be published or currently submitted for publication elsewhere. Please submit only full papers intended for review, not abstracts, to the ScholarOne portal.
Please contact the guest editors at email@example.com.
- Donatello Conte, Université de Tours, Laboratoire d’Informatique Fondamentale et Appliquée de Tours, France
- Alessandro D’Amelio, Università degli Studi di Milano, Dipartimento di Informatica, Italy
- Raffaella Lanzarotti, Università degli Studi di Milano, Dipartimento di Informatica, Italy
- Jianyi Lin, Università Cattolica del Sacro Cuore, Dipartimento di Scienze Statistiche, Italy
- Jean-Yves Ramel, Université de Tours, Laboratoire d’Informatique Fondamentale et Appliquée de Tours, France
Corresponding TETC editor: Ronald DeMara, University of Central Florida, USA (IEEE Senior Member)