Evolving a learning algorithm for the binary perceptron
José F. Fontanari, Ron Meir · Network Computation in Neural Systems · 1991
We employ a genetic algorithm to explore the space of learning algorithms for a single-layer, binary-weight neural network. We find that the directed drift algorithm, proposed recently by Venkatesh, emerges spontaneously from the genetic tournaments as the local and on-line algorithm that performs best in highly diverse environments.