An on-line learning algorithm for complex fuzzy logic
Sara Aghakhani, Scott Dick · 2010
We report on the development of an on-line learning algorithm for ANCFIS, a neuro-fuzzy architecture employing complex fuzzy sets. ANCFIS uses a hybrid learning rule, with rule consequent parameters determined by least-squares estimation in the forward pass, and premise parameters determined by a combination of gradient descent and chaotic simulated annealing in the backward pass. Our online learning algorithm replaces these with recursive least-squares in the forward pass, and the downhill-simplex algorithm in the backward pass. Experimental results on two time-series datasets show that this technique is comparable to existing results, although slightly inferior to the off-line ANCFIS results.