Graph Transduction Games Sample Clauses

Graph Transduction Games. Unlike supervised learning methods, semi-supervised learn- ing ones perform classification by leveraging the information coming from both the labeled and unlabeled observations. Their strength appears evident in those cases where the labeled set is not statistically significant to train an accurate fully- supervised classifier. In particular, the sub-category of Graph- Transductive (GT) algorithms perform the classification task in a ”transductive reasoning” fashion: rather than trying to infer an inductive, general classification rule, GT methods focus just on the observations at hand (cf. [17] for a more detailed analysis). This lack of generalization can be easily overcome through the training of a supervised classifier with the newly labeled dataset (cf. [3]). In GT the data is modeled as a graph G = (V, E, w) whose vertices are the observations in a dataset and edges represent similarities among them. V = L U , where L = (f1, y1), ..., (fl, yl) is the set of labeled observations, while U = fl+1, ..., fn is the set of unlabeled ones, with fi being feature vectors and yi 1, 2..., m being labels. Finally, w a general measure of consistency among a pair of observations assigned to each edge (u, v) E, which can be given in advance or computed using the features of the observations. ∈ { } ∈ { }