Developing an Integrated Approach to Creating a Safe and Effective Collaborative Learning Environment

Abstract

Federated Learning (FL) is an approach that makes it possible to train machine learning models on distributed data without centralizing them. Currently, there is an increasing use of FL in sensitive data areas such as medicine, finance, and IoT, where data privacy, integrity, and accessibility are critical. The purpose of the study is to systematize security threats in FL, such as attacks on confidentiality (e.g., data recovery from gradients), integrity attacks (e.g., poisoning attacks), and attacks on accessibility (e.g., denial of service). The article offers a comprehensive approach to creating a secure FL environment, including an optimized gradient encryption algorithm, an anomaly detection mechanism for detecting poisoning attacks, and a hybrid architecture combining differential privacy and secure aggregation.