Privacy-Preserving Weight Reconstruction in Federated Learning: A Legal Governance Framework Based on Algorithmic Justice

This study takes the "right to be forgotten" established by the European Union's General Data Protection Regulation (GDPR) and the compliance requirements of China's Personal Information Protection Law as its theoretical basis. It comprehensively employs a cross-disciplinary research method combining normative analysis and technical deconstruction to deeply explore the legal dilemmas inherent in the technical architecture of federated learning. The study reveals that in the operation of the federated learning system, there is a profound legal conflict between "algorithmic shadowing" and data control rights. Traditional privacy rights theories face increasing limitations in their explanatory power when confronted with this new data processing model. Based on this, this paper innovatively proposes a "dynamic responsibility allocation" legal governance paradigm. By precisely defining the rights and obligations of various participants in the federated learning ecosystem, such as data providers, model trainers, and platform operators, this paper constructs a legal governance framework that meets the requirements of algorithmic justice. The framework aims to achieve a dynamic balance between technological innovation and privacy protection, providing a solid theoretical foundation and feasible institutional guidance for the compliant and robust development of federated learning technology.