Daniel@0: Daniel@0:
Daniel@0:Daniel@0: demev2Daniel@0: Daniel@0: Daniel@0:
x
is sampled
Daniel@0: from a mixture of four Gaussians. Each class is
Daniel@0: associated with two of the Gaussians so that the optimal decision
Daniel@0: boundary is non-linear.
Daniel@0: A 2-layer
Daniel@0: network with logistic outputs is trained by minimizing the cross-entropy
Daniel@0: error function with isotroipc Gaussian regularizer (one hyperparameter for
Daniel@0: each of the four standard weight groups), using the scaled
Daniel@0: conjugate gradient optimizer. The hyperparameter vectors alpha
and
Daniel@0: beta
are re-estimated using the function evidence
. A graph
Daniel@0: is plotted of the optimal, regularised, and unregularised decision
Daniel@0: boundaries. A further plot of the moderated versus unmoderated contours
Daniel@0: is generated.
Daniel@0:
Daniel@0: evidence
, mlp
, scg
, demard
, demmlp2
Copyright (c) Ian T Nabney (1996-9) Daniel@0: Daniel@0: Daniel@0: Daniel@0: