A Single-Setup Secure Federated Learning Aggregation Protocol with Forward and Backward Secrecy for Dynamic Users

View a PDF file from the paper entitled Setting once, always insurance: one safe, safe educational assembly protocol with secure and backward for dynamic users, by Nazatuul Haque Sultan and 12 other author

PDF HTML (experimental) view

a summary:Federated Learning (FL) enables many users to cope with the automatic learning model without sharing raw data, making it suitable for privacy -sensitive applications. However, local model updates or weight are still leaking sensitive information. The safe assembly protocols reduce these risks by ensuring only the detection of collected updates. Among these, single delivery protocols are, where the generation of keys and exchange occurs only once, the most efficient due to low communications and account. However, one group protocols often lack support for the user’s dynamic participation and does not provide strong privacy guarantees such as the front and back. \ PAR in this paper, we offer a new safe assembly protocol only requires one preparation for the entire FL training. Our protocol supports the user’s dynamic participation, tolerates the dropouts, and achieves secrecy forward and back. It enhances a similar similar encryption with a major exile technology to efficiently hide updates, eliminating the need for a communication from users. To defend the typical inconsistency attacks, we offer a low -head verification mechanism using MACS authentication codes. We provide formal security evidence under each of the semi -malicious and harmful numerical models and implement a full preliminary model. Experimental results show that our protocol reduces the user’s side account of up to $ 99 % compared to modern protocols such as E-SEAFL (ACSAC’24), while maintaining the accuracy of the competitive model. These features make our protocol very practical for FL flyers in the real world, especially on resource restrictions.