Hi I am working in a project where in I develop recurrent networks using a neuro-evolution approach. I know that the Real Time Recurrent Learning algorithm by William and Zipser is only applicable to fully-connected recurrent networks. However, in my technique I will end up with a number of networks that wouldn't be fully connected. Can someone tell me if there are some gradient following algorithms that can be used to locally optimise real time recurrent networks that aren't fully connected. thanks in advance Sidhant