A learning method of a learning model for controlling an internal combustion engine, the learning method causes a processor to function so as to output a predicted value of an output parameter used for control of an internal combustion engine if actually measured values of input parameters are input,
- wherein the learning model is learned by a gradient method using a combination of actually measured values of the input parameters and an actually measured value of the output parameter as teacher data, and,
- in the learning, a learning rate adjusted to become a smaller learning rate is used when the amount of noise superposed on the actually measured value of at least one parameter among the input parameters and the output parameter is relatively large compared with when the amount of noise superposed on the actually measured value of that parameter is relatively small.