Use of Architecture-Altering Operations to Dynamically Adapt a Three-Way Analog Source Identification Circuit to Accommodate a New Source

The problem of source identification involves correctly classifying an incoming signal into a category that identifies the signal's source. The problem is difficult because information is not provided concerning each source's distinguishing characteristics and because successive signals from the same source differ. The source identification problem can be made more difficult by dynamically changing the repertoire of sources while the problem is being solved. We used genetic programming to evolve both the topology and the sizing (numerical values) for each component of an analog electrical circuit that can correctly classify an incoming analog electrical signal into three categories. Then, the repertoire of sources was dynamically changed by adding a new source during the run. The paper describes how the architecture-altering operations enabled genetic programming to adapt, during the run, to the changed environment. Specifically, a three-way source identification circuit was evolved and then adapted into a four-way classifier, during the run, thereby successfully handling the additional new source.