An event-driven neuro-fuzzy model for adaptive prognosis in homeostatic systems

This paper describes recent progress in an event-driven dynamic recurrent neuro-fuzzy model that is designed to estimate and predict states of interest within the human body. Four layers are implemented in this system, each of which consists of clusters of neurons: input layer, rule-state layer, output layer, and outcome layer. Detected events are mapped as fuzzy variables in input layer by different membership functions. The rule layer is composed of dynamic neurons, which associate with given rules. The states of a rule-neuron are not only a function of the fuzzy rule, but also on a temporal dynamic process that depends on the homeostasis, and weakly on connections with other rule-neurons that are complementary (excitatory connections) or competitive (inhibitory connections). For homeostasis, this model uses a negative feedback adaptive control system with nonlinear blocks. Sensitivity analysis and optimization tools are available to support use of the model