Fast k-Fuzzy Rough Cognitive Networks

Abstract

Fuzzy-Rough Cognitive Networks (FRCNs) are neural networks that utilize rough information granules with soft boundaries to perform the classification process. Unlike other neuronal systems, FRCNs are lazy learners in the sense that we can build the whole model while classifying a new instance. This is possible because the weight matrix connecting the neurons is prescriptively programmed. Similar to other lazy learners, the processing time of FRCN notably increases with the number of instances in the training set, while their performance deteriorates in noisy environments. Aiming at coping with these issues, this paper presents a new FRCN-based algorithm termed Fast k-Fuzzy-Rough Cognitive Network. This variant implements a parallel approach of building the information granules as computed by k-fuzzy-rough sets. Numerical simulations on 35 classification datasets show a notable reduction in the processing time of the algorithm, while delivering competitive results when compared to other lazy learners in noisy environments.