The effectiveness of evolutionary learning depends both on the structure-function relations of the organization and to variation and selection search operations applied to them. Some organizations - in particular, biological ones - are more structure-function plastic than others. We describe an abstract neuromolecular neuron model (called the cytomatrix neuron) that illustrates the structure-function plasticity necessary for evolutionary adaptability. The cytomatrix neuron is a softened cellular automaton with molecular components, roughly motivated by interactions that could occur in a molecular or cellular complex. Multiple parameters open to variation-selection. The adaptation procedure is mediated by a multiparameter evolutionary algorithm that acts on the various parameters. Experiments with both structured and unstructured learning tasks, and with difficult parity problems, demonstrate that opening more parameters to evolution increases the flexibility of generalization.
|Title of host publication||Proceedings of the 2003 International Conference on Artificial Intelligence|
|State||Published - 2003|