The University of Sussex

Using neural networks to model conditional variate densities

Peter M. Williams

Neural network outputs are interpreted as parameters of statistical distributions. This allows us to fit conditional distributions in which the parameters depend on the inputs to the network. We exploit this in modelling multivariate data, including the univariate case, in which there may be input-dependent (e.g. time-dependent) correlations between output components. This provides a novel way of modelling conditional correlation as well as providing input-dependent (local) error bars.


Download compressed postscript file