A new approach to neural networks for character recognition
Doan Bang Hoang, T.J. Jones · 2002
The paper explores a new way of creating neural networks. The central idea is that concepts are created for reasons related to the problem domain, in this case a character recognition problem, not for the reduction of an error function. The resulting system consists of a network which can be explained as a function of its learning rule, and hence the mapping attained can be justified.>