Neural network training method, system and device and medium

The invention discloses a neural network training method. The method comprises the following steps: collecting an initial search range corresponding to each hyper-parameter; searching a corresponding hyper-parameter in each initial search range by using a preset search strategy to obtain a plurality of hyper-parameter combinations; training a model by utilizing the multiple groups of hyper-parameter combinations, and screening out a plurality of hyper-parameter combinations according to a training result; determining a search range corresponding to each hyper-parameter according to the plurality of hyper-parameter combinations, and searching in each search range again by using the preset search strategy so as to obtain a plurality of hyper-parameter combinations again; training the model by using the multiple groups of hyper-parameter combinations obtained again, and screening out a final hyper-parameter combination according to a training result; and training the model by using the final hyper-parameter comb