Federated Learning with the Choquet Integral as Aggregation Method

Federated learning is a collaborative approach that enables multiple clients to jointly train a machine learning model while preserving data privacy by not sharing local data. This methodology is pivotal when individual organizations lack either the quantity or quality of data necessary to develop a robust model, particularly in dynamic environments. Our focus in this study is on horizontal federated learning. In this model, each participating client (organization) iteratively refines their model. This refinement is periodically aggregated and distributed among all members of the federation to further enhance the model's performance. Typically, the aggregation process employs a weighted average, with various methods employed to construct weights. This paper explores federated learning in the context of uncertainty and Choquet integral as an information fusion method. Additionally, we analyze the parameters of local models, which are contingent upon the efficacy of these models.