Privacy-preserving distributed learning with chaotic maps

Federated learning is a distributed machine learning approach that allows several participants to train collaboratively a machine learning model without the data leaving the participant's premises. Nevertheless there are still risks associated to the privacy of the data. In this research the authors develop a framework for training a federated Fuzzy Cognitive Map with an additional privacy layer in the form of differential privacy or chaotic maps-based encryption. The experimental tests show that there is no loss of performance by adding either privacy-preserving method.