This thesis proposes methods for modeling and learning over graphs. In the scope of graph modeling, we first address the construction of graphs suitable for compressing images that share strong redundancy — in particular, ...