Supervised and unsupervised fuzzy–adaptive Hamming net

摘要

A new fuzzy–adaptive Hamming net with supervised and unsupervised learning is proposed in this paper. The new neural model is derived from fuzzy–adaptive Hamming net and retains the advantage of deleting the search time that is a potential serious problem for ART model. The new neural model sets different vigilance parameters for different clusters (hyper-rectangles will be used in this paper). This feature makes input pattern classification more efficient when different hyper-rectangles occupy different sizes of characteristic spaces. The newly proposed supervised and unsupervised learning rule which only establishes non-cross overlapped hyper-rectangles, does not affect other hyper-rectangles during expansion and makes the classification more stable. In addition, the new learning rule allows the creation of nested hyper-rectangles in order to resolve the problem of non-convex input patterns. Simulations of the new net to palm prints recognition have been performed and good performance has been demonstrated.