CONTINUOUS BAYESIAN ESTIMATION WITH A NEURAL NETWORK ARCHITECTURE

摘要

A neural network comprises an observation system (10) sends an input observation a novum (14). The Novum (14) produces as output a process innovations suboptimal for observation and prediction entries received. prediction inputs, from an input vector (22) representing a state estimation. The output from the Novum (14) serves as an input to an infinitesimal generator (IG) (16) on an input vector (20). Said generator (IG) (16) provides state estimates on said vector (22). The novum comprises an arrangement of processing or neural elements (28), each neuron receiving said state estimates from the generator (IG) (16) on lines (32). Similarly, the generator (IG) (16) comprises a geometric neural network (34). Each neuron (34) receiving inputs from the synaptic Novum (14) on lines (36) also receives a threshold field input to amend. A quantum-wave particle is propagated to the geometrical network so as to produce an output (38) which is associated with an inertia. Each neuron (34) is associated with a memory for storing spatial models a timed series of observations. Similarly, each neuron (28) is connected to a memory for temporary storing of models of said timed series of observations. The generator (IG) (16) is adaptive and learning according to Hebbian law, while the Novum (14) is adaptive and learning according to contraHebbienne law.