Pattern Search Using Genetic Algorithms and a Neural Network Model

Abstra ct . An information processing tas k th at generates combinato rial explosion and program complexity when tr eated by a serial algorithm is investi gat ed using both genet ic algorit hms and a neur al network model. The tas k in question is to find a target memory from a set of stored entries in the form of "at t ractors" in a high-dimensional state space. The representation of entries in the memory is distributed ("an auto associat ive neur al network" in th is pap er) and t he problem is to find an attractor under a given access information where the uniqueness or even existence of a solut ion is not always gua ranteed (an ill-posed problem). The genet ic algorithm is used for generating a search orbit to search effectively for a state t hat satisfies the access condition and belongs to the target at t ra ctor bas in in the state space. The neural network is used to retri eve the corr esponding ent ry from t he networ k. T he results of our comp ut er simulations indicate that the present meth od is superior to a search meth od that uses a Mar kovian ra ndom walk in state space. Our techniques may prove useful in the realization of flexible and adapt ive informati on processing, since pattern sear ch in a high-dim ensional state space is common in various kinds of parallel informat ion process ing.