Some training theories suggest that the learning rate should be small at the beginning when network makes a lot of mistakes. If we apply strong weight changes in response to the errors, the network would experience "convulsions," avoiding some mistakes and making new ones. Early in training, a network needs a gentle and mild teacher. As it gains more knowledge, the learning rate can be increased because the network will make fewer and less serious mistakes. At the end of training after the network acquires a lot of knowledge, a slower learning rate is advisable, so that incidental mistakes (caused by a few challenges in the training set) do not spoil the result since improvement of one area produces weakening in other areas. Design experiments to confirm or negate this theory.