The inverse method for recurrent neural networks
S. Hatano, Yuji Sato, Hisaaki Hatano, T. Furuya · 2002
Investigates the inverse method for recurrent neural networks. The inverse method calculates an input of a network that locally minimizes the least-mean-square error between a given output and an output from the network. The authors compare this method with: differential algorithm, BP, bounded BP, valley searching, and bounded valley searching. The authors' experiments show that the inverse method gets better desired inputs by using bounded algorithms than by the other.>