These lectures will cover modern techniques in signal processing for data which are distributed over networks.
Analysis of data over networks or graphs (also called graph signals) is currently a very active research domain, and we will study approaches combining signal and image processing, graph theory (especially algebraic methods such as spectral analysis of graphs), and distributed methods on networks.
The application of signal processing for networks ranges from the study of technological networks (of sensors, in telecommunication or transport) to more general complex networks such as biological or social networks.
Three main topics will be explored.
First, harmonic analysis on graphs will be discussed, going from simple Fourier transform suited to graph signals to multi-scale wavelet analysis.
Second, variational approaches on graphs will be discussed, allowing to design methods such as denoising, restoration, inference or clustering on graphs.
For this purpose, modern optimization approaches based on monotone operators will also be introduced.
Finally, distributed methods will be considered, especially for consensus, estimation or detection in a network, including methods using distributed optimization.