Soft Competitive Algorithms for Learning Vector Quantization
学习矢量量化的软竞争算法

Though the loss factors in FALVQ algorithms are defined to fuzzy membership functions, the performance of the algorithms is not stable due to their scaling functions not being fuzzy membership functions. In this paper, a new family of fuzzy algorithms for the generalized LVQ network, called soft competitive algorithms for LVQ (SCALVQ), is derived from extending the definition of the loss factor corresponding to the winning prototype in FALVQ. Meanwhile, three concrete types of SCALVQ are given. In SCALVQ, the loss factors and the corresponding scaling function are both fuzzy membership functions,but an identical fuzzy membership function.Therefore,they absorb advantages of FALVQ and the soft-competition scheme,and overcome the disadvantages of the FALVQ.