Method and device for determining model structure and electronic system

The invention provides a method and device for determining a model structure and an electronic system, and the method comprises the steps: firstly extracting network layer parameters of a current network layer for each network layer in a super network; then, based on the network layer parameters, determining a probability value of each to-be-selected item in the current network layer; determininga target to-be-selected item of the current network layer from at least one to-be-selected item in the current network layer; and finally, training the super network according to the target to-be-selected item of each network layer, and determining a model structure from the trained super network. The method determines, based on network layer parameters of each network layer., the probability value of each to-be-selected item in the network layer, the better the performance of the to-be-selected item is, the larger the corresponding probability value of the to-be-selected item is, and the higher the sampled probabilit