PRUNING THRESHOLD DETERMINATION METHOD AND DEVICE, AS WELL AS MODEL PRUNING METHOD AND DEVICE

摘要

Provided by the present invention are a pruning threshold determination method and device, as well as a model pruning method and device, the pruning threshold determination method comprising: determining, for a current convolution layer of a preset model, at least one convolution kernel index position corresponding to a convolution kernel combination of the current convolution layer, wherein the current convolution layer is any convolution layer of the preset model; obtaining a cumulative distribution function according to a weight value at each convolution kernel index position; and using a preset model compression ratio as a dependent variable to substitute into the cumulative distribution function, and determining an obtained numerical value as a pruning threshold of the current convolution layer. According to the present solution, thresholds applicable to any layer of the model are determined on the basis of weight distribution conditions of said layer, thereby facilitating pruning effect optimization.