First- and Second-Order Methods for Learning: Between Steepest Descent and Newton's Method
On-line first-order backpropagation is sufficiently fast and effective for many large-scale classification problems but for very high precision mappings, batch processing may be...