On programming an adaptable network of molecular processing elements

To utilize the computational style and massive parallelism provided by molecular computing devices, conventional techniques which individually program molecules need to be augmented by techniques which collectively program a network of molecules. The author describes such an evolutionary programming technique. It consists of a training phase, in which the network is trained to recognize known patterns, and a performing phase, in which the network is presented with unknown patterns. A description is given of the results obtained when the technique is applied to a simulated network of computing devices. The computational tasks performed are those which have been traditionally difficult for conventional programming techniques.<>