Learning and Generalization in Random Automata Networks

摘要: It has been shown [7,6] that feedforward Boolean networks can learn to perform specific simple tasks and generalize well if only a subset of the learning examples is provided for learning. Here, we extend this body work show experimentally random (RBNs), where both interconnections transfer functions are chosen at initially, be evolved by using state-topology evolution solve tasks. We measure generalization performance, investigate influence average node connectivity K, system size N, introduce new allows better describe network’s behavior. Our results with higher K (supercritical) achieve memorization partial generalization. However, near critical connectivity, perfect on even-odd task.