Research
We study complex networks from complementary perspectives: their organisation across scales, their response to disruption, how useful representations can be learned from graph data and how learning can support difficult graph algorithms.
Network Robustness and Resilience How can we measure the response of a complex network to failures and targeted interventions, and identify the structures that make it fragile or resilient?
Open the permanent topic pageResearch focus
Failures and targeted interventions can change the connectivity and function of a complex system. We study theoretical and computational methods for describing that response, with particular attention to network dismantling: identifying a small set of nodes or edges whose removal fragments a network.
This work includes comparisons of dismantling methods, machine-learning approaches to structural intervention and indicators that anticipate network disintegration.
Selected publications
- Robustness and resilience of complex networks, Nature Reviews Physics (2024)
- Machine learning dismantling and early-warning signals of disintegration in complex systems, Nature Communications (2021)
- Latent Geometry-Driven Network Automata for Complex Network Dismantling, ICLR (2026)
Code and data
The network dismantling review resource collects implementations, benchmark networks and reproducibility material associated with the 2024 review.
Multilayer Networks and Mesoscale Structure How can communities, flows and other intermediate-scale structures be identified when a system contains several kinds of interaction?
Open the permanent topic pageResearch focus
Many systems contain several kinds of interaction and organise themselves at scales between individual nodes and the network as a whole. Multilayer models keep different relations distinct; community-detection and flow-based methods help reveal the resulting mesoscale organisation.
Our work has examined the community structure of multilayer systems, molecular networks, international trade and global value chains. The common question is which structural information is lost when heterogeneous relations are reduced to a single graph.
Molecular networks and biological function
Our study of the human proteome, with Giuseppe Jurman and Manlio De Domenico, keeps molecular interaction layers separate when identifying functional modules with Multiplex Infomap. Its random walker follows the links of its current layer and, with a probability called the relax rate, the links of any layer, which lets it move between layers. An information-theoretic criterion selects this rate.
Comparing the resulting modules with independently annotated gene sets assesses their biological relevance. The selected multilayer organisation has greater functional content than modules obtained from the aggregate representation, connecting a modelling choice to an external biological assessment.
Selected publications
- The nested structural organization of the worldwide trade multi-layer network, Scientific Reports (2019)
- Modularity in multilayer networks using redundancy-based resolution and projection-based inter-layer coupling, IEEE Transactions on Network Science and Engineering (2019)
- Multilayer flows in molecular networks identify biological modules in the human proteome, IEEE Transactions on Network Science and Engineering (2020)
Learning on Graphs How can machine-learning models use relational structure without discarding edge types, weights, signs, layers or other information carried by a graph?
Open the permanent topic pageResearch focus
Graph data are rarely homogeneous. Relations may be weighted, signed, organised in layers or observed only indirectly. We study learning methods that preserve and exploit this structure rather than flattening it before analysis.
This line includes graph neural networks for non-standard graph structures, link prediction, graph generation and methods for evaluating or interpreting learned network representations. Besides predictive performance, we look for models whose assumptions and outputs can be related back to network structure.
Selected publications
- wsGAT: Weighted and Signed Graph Attention Networks for Link Prediction, Complex Networks (2021)
- mGNN: Generalizing the Graph Neural Networks to the Multilayer Case, arXiv preprint (2021)
- Beyond MMD: Evaluating Graph Generative Models with Geometric Deep Learning, arXiv preprint (2025)
- ProtoGuide: Prototype-Driven Guidance for Class-Conditional Graph Generation, arXiv preprint (2026)
Graph Algorithms and Combinatorial Optimisation How can structural information and machine learning make hard graph problems more tractable while preserving clear objectives and measurable performance?
Open the permanent topic pageResearch focus
Many questions about complex networks lead to computationally difficult graph problems. We investigate hybrid approaches in which a solver stays in charge and learning helps it decide where to search, which elements to prune or which intervention to evaluate next.
Our papers cover search-space reduction for maximum-clique problems and node- or edge-based network dismantling. Across these settings, we are interested in scalability, comparison against strong algorithmic baselines and what the learned strategy reveals about the structure of the problem.
Selected publications
- Learning fine-grained search space pruning and heuristics for combinatorial optimization, Journal of Heuristics (2023)
- Geometric deep learning sub-network extraction for maximum clique enumeration, PLOS ONE (2024)
- Edge Dismantling with Geometric Reinforcement Learning, Complex Networks (2024)