Activity level of a neural net and its learning environment

Summary form only given, as follows. When a neural net is used to solve continuous problems, the learning environment, which may influence convergence and accuracy, differs from that for true-false problems. Based on the energy model for a neural net, different activity levels of the net are generalized to learn one selected continuous problem-polynomial function. The training results showed that there are some optimal activity levels that lead the net to obtain better accuracy than that from other levels. The concepts of maximum energy and minimum energy (or `thermal noise´) are proposed to explain why it is possible for a net to achieve a good learning environment to fit to the continuous problems