Federal learning method and system based on generalized linear regression, terminal and medium

The invention provides a federated learning method and system based on generalized linear regression, a terminal and a medium, and the method comprises the steps: two data parties jointly construct a generalized linear regression model, and the two data parties comprise a data holder A end for providing a training attribute value and a data application side B end for providing a data label value; a loss function of the generalized linear regression model is constructed, and when variables of the generalized linear regression model meet Tweedie distribution, the training target is the minimum loss function; the data holder A end solves partial derivative according to the loss function to obtain a local gradient, and exchanges an intermediate result used for calculating the local gradient through a homomorphic encryption technology; and the data holder A end updates the generalized linear regression model by using a local gradient. The invention provides a two-party longitudinal federated learning scheme of a g