Computing OWA weights as relevance factors

* On leave of absence from the Department of Electronics and Computers, Transylvania University of Braşov. Abstract – Ordered Weighted Aggregation (OWA) operators represent a distinct family of aggregation operators and were introduced by Yager in [1]. They compute a weighted sum of a number of criteria that must be satisfied. The central element of the OWA operators is that the criteria are reordered before aggregation and therefore a particular weight is associated to a position. Relevance Learning Vector Quantization (RLVQ) [2] is an extension of the Learning Vector Quantization (LVQ) algorithm [3] and performs a heuristic determination of the relevance factors of the input dimensions. This method is based on Hebbian learning and associates a weight factor to each dimension of the input vectors. We present a LVQ method for on-line computing of the OWA weights as relevance factors. The method uses a weighted metric based on OWA restrictions. The principal benefit of our algorithm is that it connects two distinct topics: RLVQ algorithm and the consistent mathematical model of the OWA operators.