Lambda learning rule for feedforward neural networks
Jacek M. Żurada · 2002
Feedforward layered networks of continuous perceptrons are traditionally trained using the delta and generalized delta training rules. These learning concepts are formalized in the error backpropagation training (EBPT) concept. Although the EBPT algorithm is widely used, the lambda learning rule often offers a considerable improvement in learning. Both the rule and the complete generalized lambda learning algorithm for layered networks are outlined. Emphasis is placed on visualization of learning. Comparisons between the two learning approaches are drawn.>