GenNets: genetically programmed neural nets-using the geneticalgorithm to train neural nets whose inputs and/or outputs vary in time

摘要

The author shows that the generic algorithm (GA) can be appliedsuccessfully to training nonconvergent networks, and presents someexamples of their extraordinary behavioral versatility. He first gives abrief summary of the GA and the genetic programming of neural networks.He shows how GP techniques were used to evolve GenNets with specifiedoperating conditions, and demonstrates some of the extraordinarycapacities of time-dependent GenNets. He also makes a plea to the neuralnetwork research community to `shift its sights upwards' by devotingmore effort to thinking about `dynamic' neural networks in general, andthe theory of GenNet dynamics and `evolvability' in particular