Neural network learning in an ecological and evolutionary context

Neural network models are inspired by the nervous system of real organisms but these organisms are usually viewed as living in an ecological and evolutionary void. By assuming instead that neural networks control the behavior of organisms in an environment and are part of evolving populations subject to selective reproduction and mutation, learning can be studied in these networks in ways which are different from those of typical connectionist simulations of learning. We describe simulations in which learning occurs in an ecological and evolutionary context and we show how this type of learning has implications for what is learned, the control of the learning experience (i.e. the sequence of inputs during learning) by the network itself, initial conditions, and teaching input.