Federal learning system using data summary and method thereof

A federated learning method using a data summary includes a coordinating device sending a generic model to a plurality of client devices, each client device generating an encoding feature according to raw data and executing a training program, the training program including: updating the generic model to generate a client model, when a summary request is received, generating a data summary; and selecting at least two coding features and at least two labels to calculate a feature weighted sum and a label weighted sum, and sending the feature weighted sum and the label weighted sum as an abstract to the coordination device together with update parameters of the customer model, and the coordination device executes the following steps: judging an absent customer and an attended customer in the customer device; generating an alternative model according to the general model and the absent customer, generating an aggregation model according to the customer model of the attended customer and the alternative model, an