| |||||||||||
|
The Graph Neural Network model The Graph Neural Network (GNN) is a novel connectionist model particularly suited for problems whose domain can be represented by a set of patterns and relationships between them [1,2]. In those problems, a prediction about a given pattern can be carried out exploiting all the related information, which includes the pattern features, the pattern relationships and, in general, the whole graph that represents the domain. GNN peculiarity consists in its capability of making the prediction taking directly in input the domain graph, without any preprocessing. In this sense, the GNN methods is different from the common approach, which face a domain with relationship by an ad hoc preprocessing procedure that compresses into a vectorial representation all the data about a pattern. Actually, GNNs can be considered the connectionist counterpart of SVM for graphs and random fields. GNNs have been proved to be sort of universal approximator for functions on graphs and have been applied to several problems, including spam detection, object localization in images, molecule classification. The GNN software The GNN simulator has been implemented in Matlab and written by me and Gabriele Monfardini. The toolbox, along with benchamarks used in [1,2], and a short manual is avalable for download in the following. References
|
||||||||||
