Evolutionary graph classification systems by granular computing based embedding

Graphs have gained a lot of attention in the pattern recognition community thanks to their ability to encode both topological and semantic information. Despite their invaluable descriptive power, their arbitrarily complex structured nature poses serious challenges when they are involved in learning systems. Typical approaches aim at building a vectorial representation of the graph in a suitable embedding space by leveraging on the selection of relevant prototypes that enable the use of common pattern recognition methods. An interesting paradigm able to synthesize prototypes in a data-driven fashion can be found in Granular Computing. This thesis investigates and develops novel techniques for graph embedding methods based on Granular Computing and Multi-Agent based systems in order to solve Pattern Recognition problems. Initially, the proposed methods aim at improving different aspects of an established Granular Computing-based framework designed for graph classification concerning the computational complexity, granulation and embedding ability and optimization problem of the training phase. A lightweight stochastic procedure for the selection of prototypes has been designed in order to mitigate the computational burden of the algorithm. Other proposed techniques focus on improving the granulation phase of the framework by selecting detailed granules of information that characterize the classes of the problem at hand. Concerning the embedding phase, six different graph embedding techniques inspired by the dissimilarity space embedding are proposed to represent graphs into meaningful embedding spaces. Concerning the optimization phase, a novel evolutionary-based approach has been designed for equipping the framework with a class-specific metric learning strategy together with a reformulation in a multi-objective fashion of the problem which aims at jointly optimizing the performance of the classifier, the number of information granules and the structural complexity of the classification model. In the second part ...