Lightweight Federated Learning for Secure and GDPR-Compliant Elderly Care: Lightweight Federated Learning and Cross-Modal Game Architecture

Escalating security threats in integrated elderly care systems necessitate architectures that simultaneously ensure data integrity, privacy preservation, and multi-party collaboration. This study proposes a novel security framework integrating three innovations: (1) A response entropy optimization model dynamically quantifies system vulnerability and adjusts defense strategies in real-time; (2) Lightweight federated learning with blockchain-based notarization achieves 100% raw data localization through gradient compression (62% overhead reduction) and zero-trust access control; (3) A cross-modal game architecture resolves medical-insurance-civil conflicts via Nash equilibrium solutions, reducing policy execution latency to 0.5s. Validated across 104 institutions (2,300+ beds), the architecture intercepted 237 data breaches, cut down privacy risks by 92%, and accelerated illegal access response to 8ms. This work establishes a replicable paradigm for secure elderly care that balances GDPR compliance with operational efficiency.