wolffd@0: prior = mlpprior(nin, nhidden, nout, aw1, ab1, aw2, ab2)wolffd@0: wolffd@0: wolffd@0:
prior = mlpprior(nin, nhidden, nout, aw1, ab1, aw2, ab2)
wolffd@0: generates a data structure
wolffd@0: prior, with fields prior.alpha and prior.index, which
wolffd@0: specifies a Gaussian prior distribution for the network weights in a
wolffd@0: two-layer feedforward network. Two different cases are possible. In
wolffd@0: the first case, aw1, ab1, aw2 and ab2 are all
wolffd@0: scalars and represent the regularization coefficients for four groups
wolffd@0: of parameters in the network corresponding to first-layer weights,
wolffd@0: first-layer biases, second-layer weights, and second-layer biases
wolffd@0: respectively. Then prior.alpha represents a column vector of
wolffd@0: length 4 containing the parameters, and prior.index is a matrix
wolffd@0: specifying which weights belong in each group. Each column has one
wolffd@0: element for each weight in the matrix, using the standard ordering as
wolffd@0: defined in mlppak, and each element is 1 or 0 according to
wolffd@0: whether the weight is a member of the corresponding group or not. In
wolffd@0: the second case the parameter aw1 is a vector of length equal to
wolffd@0: the number of inputs in the network, and the corresponding matrix
wolffd@0: prior.index now partitions the first-layer weights into groups
wolffd@0: corresponding to the weights fanning out of each input unit. This
wolffd@0: prior is appropriate for the technique of automatic relevance
wolffd@0: determination.
wolffd@0:
wolffd@0: mlp, mlperr, mlpgrad, evidenceCopyright (c) Ian T Nabney (1996-9) wolffd@0: wolffd@0: wolffd@0: wolffd@0: