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