wolffd@0: net = mdn(nin, nhidden, ncentres, dimtarget) wolffd@0: net = mdn(nin, nhidden, ncentres, dimtarget, mixtype, ... wolffd@0: prior, beta) wolffd@0:wolffd@0: wolffd@0: wolffd@0:
net = mdn(nin, nhidden, ncentres, dimtarget) takes the number of
wolffd@0: inputs,
wolffd@0: hidden units for a 2-layer feed-forward
wolffd@0: network and the number of centres and target dimension for the
wolffd@0: mixture model whose parameters are set from the outputs of the neural network.
wolffd@0: The fifth argument mixtype is used to define the type of mixture
wolffd@0: model. (Currently there is only one type supported: a mixture of Gaussians with
wolffd@0: a single covariance parameter for each component.) For this model,
wolffd@0: the mixture coefficients are computed from a group of softmax outputs,
wolffd@0: the centres are equal to a group of linear outputs, and the variances are
wolffd@0: obtained by applying the exponential function to a third group of outputs.
wolffd@0:
wolffd@0: The network is initialised by a call to mlp, and the arguments
wolffd@0: prior, and beta have the same role as for that function.
wolffd@0: Weight initialisation uses the Matlab function randn
wolffd@0: and so the seed for the random weight initialization can be
wolffd@0: set using randn('state', s) where s is the seed value.
wolffd@0: A specialised data structure (rather than gmm)
wolffd@0: is used for the mixture model outputs to improve
wolffd@0: the efficiency of error and gradient calculations in network training.
wolffd@0: The fields are described in mdnfwd where they are set up.
wolffd@0:
wolffd@0:
The fields in net are
wolffd@0:
wolffd@0: wolffd@0: type = 'mdn' wolffd@0: nin = number of input variables wolffd@0: nout = dimension of target space (not number of network outputs) wolffd@0: nwts = total number of weights and biases wolffd@0: mdnmixes = data structure for mixture model output wolffd@0: mlp = data structure for MLP network wolffd@0:wolffd@0: wolffd@0: wolffd@0:
wolffd@0: wolffd@0: net = mdn(2, 4, 3, 1, 'spherical'); wolffd@0:wolffd@0: wolffd@0: This creates a Mixture Density Network with 2 inputs and 4 hidden units. wolffd@0: The mixture model has 3 components and the target space has dimension 1. wolffd@0: wolffd@0:
mdnfwd, mdnerr, mdn2gmm, mdngrad, mdnpak, mdnunpak, mlpCopyright (c) Ian T Nabney (1996-9) wolffd@0:
David J Evans (1998) wolffd@0: wolffd@0: wolffd@0: