CONTINUOUS BAYESIAN ESTIMATION WITH A NEURAL NETWORK ARCHITECTURE

摘要

Nerve (10) output includes new (14) of observation input. New (14) provide the innovation process of output suboptimum in relation to received observation and receive prediction input. The received prediction of institute inputs received input vector (22) and represents a state estimation. The newly output of (14) is input into infinitesimal generator (IG) (16) to input vector (20). The IG (16) provides the vector (22) of state estimation. This newly includes processing element array or neuron (28), each receives state and estimates from above-mentioned IG (16) in route (32). In a similar fashion, IG (16) is by geometry dot matrix neuron (34). Each neuron (34) receives synaptic input and new (14) are online (36), and also receives the input of modification threshold field. It is associated with inertia to provide output (38) in geometric grid that quanta particle propagates mechanical wave. The IG (16) of each neuron (34) has associated memory, for storing the time sight value ordered series of numbers of the space pattern, in a similar way, neuron (28) respectively has an associated memory, for storing temporal pattern observation ordered series of numbers at that time. This is adaptive and has learned IG (16) and thus new (14) is adaptive and learned by contraHebbian rules by Hebbian laws.