Federal data analysis system and method based on gradient dynamic cutting

The invention discloses a federal data analysis system and a federal data analysis method based on gradient dynamic cutting, which cut a gradient which is uploaded to a server by a client and contains a certain noise through a technology of dynamically generating a cutting threshold, and improve the precision of a model on the premise of not reducing privacy protection. As no prior boundary exists, the influence difference of noise on the gradient is large, and the accuracy is reduced. Therefore, the accuracy of the training model can be well improved by setting the proper clipping threshold. The method comprises the following steps: firstly, constructing a dynamic cutting threshold function according to the actual condition and cutting loss of gradient norm change of each round of iteration, and adjusting a gradient cutting threshold; and based on the overall trend of the overall change of the gradient norm, calculating a gradient attenuation clipping threshold fitting curvature, and adjusting a dynamic clip