Missing and noisy data in nonlinear time-series prediction.

Tresp, Volker , Reimar Hofmann
Missing and noisy data in nonlinear time-series prediction.
Neural Networks for Signal Processing 5 ( gzipped Postscript - )

Abstract: This paper is now of mostly historical importance. At the time of publication (1995) it was one of the first machine learning papers to stress the importance of stochastic sampling in time-series prediction and time-series model learning. In this paper we suggested to use Gibbs sampling, nowadays particle filters are commonly used instead. Secondly, this is one of the first papers in machine learning to derive the gradient equations for control optimization in reinforcement learning policy-space search methods. Since our paper was addressed to a neural network community, we focussed on a neural network representation with Gaussian noise.