Network (graph) Learning


Involved faculty members:

Elvin Isufi Alan Hanjalic


The increasing amount of data generated by networks, such as social, recommender systems, or water networks requires developing algorithms to learn meaningful representation by accounting for the underlying interconnections. These interconnection serve as a support for the data, which if exploited property can lead to substantial gains of artificial intelligence techniques. In the Multimedia Computing Group, we are conducting fundamental and applied research on the following interleaved directions:

  • Develop machine learning techniques to account for the underlying structure in the data. We are focusing specifically on graph neural networks, which extend deep learning techniques to network data. By leveraging the structure as a prior for the data at hand, we do not only develop learning algorithms that are data efficient and computationally light but also characterise theoretically the behaviour of such algorithms.
  • Build processing algorithms for time-varying network data. We are leveraging tools from graph signal processing to develop methods that capture data relations in a spatio-temporal manner. These tools allow us also to identify the network topology from partial and uncertain data.
  • Graph-based recommender systems. Recommender systems have intrinsic interactions between users and items, which can be modelled through a graph. Building upon these interactions we are developing graph-based recommender system algorithms to personalise recommendation.

Research in these areas is flourishing through solid collaborations with the University of Pennsylvania, University of California Berkley, EPFL, University of Agder, and La Sapienza University of Rome which have lead to several publications in elite venues. These venues include IEEE Transactions on Signal Processing, Signal Processing Magazine, ICASSP conference and Elsevier Information Processing and Management.


Representative publications

  1. E. Isufi, M. Pocchiari and A. Hanjalic, Accuracy-Diversity Trade-off in Recommender Systems via Graph Convolutions, Elsevier Information Processing and Management, Jul. 2021.
  2. F. Gama, E. Isufi, G. Leus and A. Ribeiro, From Graph Filters to Graph Neural Networks, to appear in the IEEE Signal Processing Magazine; Special Issue on Graph Signal Processing: Foundations and Emerging Directions. 2020.
  3. M. Coutino, E. Isufi, T. Maehara and G. Leus, State-Space Network Topology Identification from Partial Observations, to appear in the IEEE Transactions on Signal and Information Processing over Networks; Special Issue on Network Topology Identification, 2020.
  4. E. Isufi, A. Loukas, N. Perraudin and G. Leus, Forecasting Time Series with VARMA Recursions on Graphs, IEEE Transactions on Signal Processing, 2019.
  5. B. Iancu, L. Ruiz, A. Ribeiro and E. Isufi, Graph-Adaptive Activation Functions For Graph Neural Networks, IEEE International Workshop on Machine Learning for Signal Processing, Espoo, Finland, Sep. 2020.
  6. E. Isufi, F. Gama and A. Ribeiro, Generalizing Graph Convolutional Neural Networks with Edge Variant Recursions on Graphs, to appear in the EURASIP European Signal Processing Conference (EUSIPCO), A Coruña, Spain, Sep. 2019.
/* */