The parametric avalanche: continuous Bayesian estimation and control with a neural network architecture

Summary form only given, as follows. The parametric avalanche is a new neural network architecture which is designed to perform adaptive continuous Bayesian estimation on unpreprocessed large dimensional data. It is adaptive in the sense that it ´learns´ the infinitesimal generators of nonlinear plant trajectories through observation. Then, when presented with previously learned spatio-temporal patterns, it associatively accesses the stored dynamical equations and uses them to generate continuous estimates of the observed system´s state. By using the innovations method for stochastic estimation, the parametric avalanche automatically performs optimal data compression on the stored representations. Use of soliton wave propagation in the threshold field provides a compact representation of stored data, tracking before detection, trajectory extrapolation through loss of signal, and a host of other capabilities.<>