wolffd@0: wolffd@0:
wolffd@0:wolffd@0: demmlp1wolffd@0: wolffd@0: wolffd@0:
x1
is sampled uniformly from the range (0,1) and has
wolffd@0: a low level of added Gaussian noise, x2
is a copy of x1
wolffd@0: with a higher level of added noise, and x3
is sampled randomly
wolffd@0: from a Gaussian distribution. The single target variable is determined
wolffd@0: by sin(2*pi*x1)
with additive Gaussian noise. Thus x1
is
wolffd@0: very relevant for determining the target value, x2
is of some
wolffd@0: relevance, while x3
is irrelevant. The prior over weights is
wolffd@0: given by the ARD Gaussian prior with a separate hyper-parameter for
wolffd@0: the group of weights associated with each input. A multi-layer
wolffd@0: perceptron is trained on this data, with re-estimation of the
wolffd@0: hyper-parameters using evidence
. The final values for the
wolffd@0: hyper-parameters reflect the relative importance of the three inputs.
wolffd@0:
wolffd@0: demmlp1
, demev1
, mlp
, evidence
Copyright (c) Ian T Nabney (1996-9) wolffd@0: wolffd@0: wolffd@0: wolffd@0: