Privacy-Preserving Secure Distributed Computer Vision for Malaria Cells Detection
Irina Arévalo, Jose L. Salmeron, Ian De La Oliva · 2024
Federated learning is a machine learning approach that enables multiple participants to collaboratively train a deep learning model using their sensitive data, all while maintaining data privacy. In this research, the authors employ an additional privacy layer with chaotic-maps based encryption for computer vision. This method is compared to the use of differential privacy as privacy layer. The experimental approach evaluates the performance of both methods and the results confirm that the federated learning process improves the average performance metrics of the computer vision model for all the experiments while maintaining the process secure, and that the chaotic-maps based encryption does not worsen the results of differential privacy.