Cesare Alippi

Graph Neural Networks rely on graph structure as a fundamental inductive bias, yet in many real-world applications the underlying relational structure is unknown, incomplete, or potentially inaccurate. Graph Structure Learning (GSL) addresses this challenge by inferring relational structures directly from data and optimizing them for the downstream task. This seminar provides an overview of the main approaches to GSL, focusing on structure initialization, edge scoring, graph sparsification, and loss design. We discuss several strategies for constructing an initial graph, including Pearson correlation, Granger causality, pairwise input similarity, Dirichlet energy minimization, and graph rewiring. Different edge-scoring approaches are then considered, ranging from input-independent parameterizations to input-dependent models based on neural networks and attention mechanisms.
Particular attention is given to the challenges of learning sparse and discrete graph structures, including thresholding and Bernoulli sampling, as well as the gradient-estimation techniques required for their optimization. Finally, we discuss how downstream objectives and regularization terms can guide the learned graph toward structures that are both task-relevant and consistent with desired properties. Overall, the seminar highlights the potential of GSL as an alternative to predefined graph topologies, while emphasizing the trade-offs between expressiveness, computational efficiency, and structural constraints.

Biography:

CESARE ALIPPI is Professor with the Università della Svizzera italiana (Switzerland) and Professor with the Politecnico di Milano (Italy); he is visiting Professor at the Guandong University of Technology (China) and Consultant Professor at the Northwestern Polytechnic of Xi’An (China).

He has been a visiting researcher/professor at UCL (UK), MIT (USA), ESPCI (F), CASIA (RC), A*STAR (SIN), U.Kobe (Japan).

Alippi is an IEEE Fellow, ELLIS Fellow, INNS Fellow and AAIA Fellow, Past Board of Governors member of the International Neural Network Society, Past member of the Administrative Committee of the IEEE Computational Intelligence Society (CIS), Past Board of Directors member of the European Neural Network Society, Past Vice-President education of the IEEE Computational Intelligence Society, Associate Editor of Proceedings of IEEE and other journals, Past Associate editor of the IEEE Computational Intelligence Magazine, the IEEE-Transactions on Instrumentation and Measurements, IEEE-Transactions on Neural Networks, IEEE-Transactions on Emerging Topics in Computational Intelligence and member and chair of many IEEE committees.

In 2024 he received the IEEE CIS Enrique Ruspini Meritorious Service Award, the 2018 IEEE CIS Outstanding Computational Intelligence Magazine paper award, the 2016 Gabor Award from the International Neural Networks Society and the Outstanding Transactions on Neural Networks and Learning Systems Paper Award from the IEEE Computational Intelligence Society; in 2013 the IBM Faculty award; in 2004 the IEEE Instrumentation and Measurement Society Young Engineer Award.

Current research activity addresses graph-based learning, adaptation and learning in non-stationary environments and Intelligence for embedded, cyber-physical systems and IoT. For the graph-based learning research please refer to http://gmlg.ch

He holds 8 patents, has published one monograph book (translated in Chinese), 7 edited books and about 250 papers in international journals and conference proceedings.