Artificial Intelligence: New Frontiers for the Analysis and Understanding of Complex Networks
A collaboration between the Scuola Superiore Sant’Anna, Scuola IMT Alti Studi Lucca, and Aalborg University has led to the development of an innovative method to improve AI models for the analysis of data and complex systems.
Our everyday lives are immersed in systems built on connections: city roads, urban and suburban transport networks, and relationships on social media platforms. All these systems are examples of “complex networks”: structures in which different elements are interconnected in ways that are often difficult to analyze. Understanding how these networks function is a crucial challenge for artificial intelligence.
A new study published in IEEE Transactions on Knowledge and Data Engineering, one of the leading international journals in Artificial Intelligence, introduces an innovative approach to help AI systems represent and analyze complex networks. The research is the result of a collaboration between the Scuola Superiore Sant’Anna, Scuola IMT Alti Studi Lucca, and Aalborg University.
A New Network Embedding Method to Understand Complex Systems
Today, many types of information are represented as networks (or graphs), in which nodes and elements are connected through relationships. To allow artificial intelligence algorithms to process these data, network embedding operations are required — namely, the transformation of structured data into compact numerical representations.
The limitation of traditional methods is that they often focus only on the “distance” between elements, overlooking a crucial aspect: the role each node plays within the network.
The study proposes a new network embedding method capable of preserving not only the proximity between nodes, but also the structural role they play within the system.
Another key strength of the method is its scalability: the technique is designed to work on very large networks, including those with complex and directional connections, while maintaining significantly reduced computation times compared with existing solutions.
The study involves an interdisciplinary research team including Giuseppe Squillace, first author of the paper and researcher at Université Paris-Saclay, previously at the Scuola IMT Alti Studi Lucca; Mirco Tribastone, Full Professor at the Scuola IMT Alti Studi Lucca; Max Tschaikowski, Associate Professor at Sapienza University of Rome and formerly at Aalborg University; and Andrea Vandin, Associate Professor at the Institute of Economics and the Department of Excellence EMbeDS Department of the Scuola Superiore Sant’Anna.
According to the authors, “this work shows how preserving structural relationships between entities is crucial for improving network embeddings, proposing a new way to extract information from complex data for the benefit of artificial intelligence and machine learning techniques.”
Potential Applications
The experiments presented in the article demonstrate that the new method enables extremely efficient processing of large-scale networks across multiple application domains, including transport networks, biological systems, and social networks.
The experimental results clearly highlight the superiority of the approach compared with the current state of the art: the generated embeddings are interpretable, are computed in significantly shorter times, and the models using them achieve better performance.
Cover image generated with AI
