From AMS Glossary
The error in the output of the neural network (e.g., a prediction of minimum nighttime temperature) is a function of the values of the input parameters (cloud cover, 6
temperature, etc.) and the weights assigned to them. The weights are adjusted to minimize the error. Application of the delta rule results in the most rapid error reduction (learning rate). The learning rate can be adjusted as necessary to avoid being trapped in a local error minimum.