This paper develops PAC (probably approximately correct) error bounds for network classifiers in the transductive setting, where the network node inputs and links are all known, the training nodes class labels are known, ...