Genetic design of biologically inspired discrete dynamical basis networks to approximate data sequences

This paper described a method by which time based data sequences are modeled using a biologically inspired, discrete dynamical basis network (DDBN) of simple operators. Each operator contains an internal state variable, and the operator network represents a system of recurrence relations. The temporal values of each state variable form a basis sequence. The least mean squares (LMS) weighted sum of basis sequences compute the output from the network. A genetic algorithm (GA) was used to evolve the DDBN. Three different target data sequences were chosen to test the genetic design of the DDBNs. Results indicate that DDBNs with LMS summing output junctions are effective in approximating time based data sequences and capable of storing information in a procedural, dynamical fashion.