A novel concept for first order learning algorithm design
Peter Géczy, S. Usui · 2002
One of the essential problems in the neural network field is the fact that some learning techniques perform well on certain classes of problems and fail on the others. Conventional approaches to training neural networks overlook the important link between the learning algorithm and the learning task. Ignoring such evidence leads to various controversies. To resolve the issue requires us to establish a suitable classification framework for both learning algorithms and learning tasks.