CONTINUOUS BAYESIAN ESTIMATION WITH A NEURAL NETWORK ARCHITECTURE

摘要

A neural network includes an observation system (10) which outputs an observation input to a novum (14). The novum (14) provides on an output a suboptimal innovations process related to the received observation and received prediction inputs. The received prediction inputs are received from an input vector (22) and represent a state estimate. The output of the novum (14) is input to an infinitesimal generator (IG) (16) on input vector (20). The IG (16) provides the state estimates on the vector (22). The novum is comprised of an array of processing elements or neurons (28) which each receive the state estimates from the IG (16) on lines (32). In a similar manner, the IG (16) is comprised of a geometrical lattice of neurons (34). Each of the neurons (34) receive synaptic inputs from the novum (14) on lines (36) and also receive a modifying threshold field input. A quantum mechanical wave particle is propagated across the geometrical lattice to provide an output (38) which has an inertia associated therewith. Each of the neurons (34) in the IG (16) has associated therewith a memory for storing the spatial patterns of a timed series of observations and, in a similar manner, the neurons (28) each have a memory associated therewith for storing the temporal patterns of the timed series of observations. The IG (16) is adaptive and learns by the Hebbian law whereby the novum (14) is adaptive and learns by the contraHebbian law.