The optimised internal representation of multilayered classifier networks performs nonlinear discriminant analysis

摘要: Abstract This paper illustrates why a nonlinear adaptive feed-forward layered network with linear output units can perform well as pattern classification device. The central result is that minimising the error at of equivalent to maximising particular norm, discriminant function, hidden units. first part explicitly performing transformation data into space in which classes may be more easily separated. specific nature this constrained maximise function. If targets are appropriately chosen, function relates pseudo-inverse total covariance matrix and weighted between-class unit patterns. Numerical simulations presented illustrate results.