Fuzzy ARTMAP has been proposed as a neural network architecture for supervised learning of recognition categories and multidimensional maps in response to arbitrary sequences of analog or binary input vectors [12]. In this paper, RePART, a proposal for a variant of Fuzzy ARTMAP is analysed. As in ARTMAP-IC, this variant uses distributed code processing and instance counting in order to calculate the set of neurons used to predict untrained data. However, it additionally uses a reward/ punishment process and takes into account every neuron in the calculation process.....
展开▼
