Learning algorithm for nearest-prototype classifiers
E. Urahama, Takeshi Nagao · 2005
Incremental learning algorithms are presented for nearest prototype (NP) classifiers. Fuzzification of the 1-NP and K-NP classification rules provides an explicit analytical expression of the membership of data to categories. This expression enables formulation of the protoype placement problem as mathematical programming which can be solved by using a gradient descent algorithm. In addition to the learning algorithm, analog electronic circuits are configured, which implement the 1-NP and k-NP classifiers.