Many real networks have hidden geometric structure: nodes that are close in a latent space are more likely to interact, exchange information, or play related functional roles. I develop methods for mapping complex networks into multidimensional hyperbolic spaces, estimating their effective dimensionality, and using these maps for prediction, interpretation, and machine-learning applications.

• The D-Mercator method for the multidimensional hyperbolic embedding of real networks. Nature Communications, 2023.
• Feature-aware ultra-low dimensional reduction of real networks. npj Complexity, 2024.
• Mapping bipartite networks into multidimensional hyperbolic spaces. Communications Physics, 2026.
• Chordless cycle filtrations for dimensionality detection in complex networks via topological data analysis. Nature Communications, 2026.

I use network science to understand how AI systems learn, compute, and become robust. The goal is to treat neural networks and graph-learning systems as measurable complex systems, where task structure, topology, representations, and failure modes can be studied with quantitative tools rather than only benchmark scores.

• Task complexity shapes internal representations and robustness in neural networks. Machine Learning: Science and Technology, 2026.
• HypBench: Hyperbolic Benchmark for Graph Neural Network Performance. IEEE Transactions on Neural Networks and Learning Systems, 2026.
• Topological Deep Learning Challenge 2025: Expanding the Data Landscape. Proceedings of Machine Learning Research, 2026.

My postdoctoral work applies latent geometry and network-science methods to biological networks, especially protein-protein interaction networks. These projects mainly focus on studying the role of complementarity in protein interactions and applying latent geometry for the identification of protein pathways.