Abstract: Reading temporal networks through graph evolution rules

Temporal networks encode not only when interactions occur, but also how recurrent mesoscopic mechanisms progressively reshape connectivity. Yet, most approaches either extend static descriptors to time-stamped data or adopt growth models that pre-impose a small set of mechanisms. In this talk I present a framework to read temporal networks through Graph Evolution Rules (GERs): probabilistic patterns, inspired by association rules, in which a subgraph observed at a given time tends to evolve into one or more subgraphs at a subsequent with measurable likelihood. GERs provide an explicit, mechanism-oriented representation of network dynamics and yield a compact, interpretable Network Evolution Profile—a distribution over statistically validated rules that captures the evolution signature of a system and supports rigorous comparison across graphs, communities, and ego-networks. This perspective opens the way to mechanism-preserving GNN architectures, improved explainability grounded in explicit evolutionary transformations, and practical applications such as anomaly detection and distributional shifts framed as changes (drift or change points) in the evolution profile.

Title: Introduction to PyTorch Geometric 📐

This coding session will cover the basics of PyTorch Geometric, including data handling, model building, and training of graph neural networks. Students will get hands-on experience with key concepts and tools.

Title: Advanced PyTorch Geometric 📐 with Torch Geometric Pool 🎱

This coding session will cover advanced topics in PyTorch Geometric with the Torch Geometric Pool library, including pooling techniques and graph coarsening. Participants will learn how to implement and utilize these methods effectively to improve graph neural network performance.

Abstract: Shaping GNNs with dynamical systems

The dynamics of information diffusion in local message passing is a key issue that heavily influences graph representation learning, especially when long-range propagation is needed. We vouch for principled approaches that control and regulate the degree of propagation, conservation and dissipation of information throughout the neural flow. In the talk we explore some approaches stemming from a dynamical systems' view of neural networks for static and temporal graphs, with particular focus on wide applicability of the concepts and on the theoretical guarantees on information conservation.

Title: Graph Deep Learning for Time Series Forecasting

Graph deep learning methods have become popular tools to process collections of correlated time series. Unlike traditional multivariate forecasting methods, graph-based predictors leverage pairwise relationships by conditioning forecasts on graphs spanning the time series collection. This tutorial will explore these methods in depth, providing practical examples and insights.