The advent of Federated Learning (FL) as a distributed machine learning paradigm has introduced new cybersecurity challenges, notably adversarial attacks that threaten model integrity and participant privacy. This study proposes an innovative security framework inspired by Control-Flow Attestation (CFA) mechanisms, traditionally used in cybersecurity, to ensure software execution integrity. By integrating digital signatures and cryptographic hashing within the FL framework, we authenticate and verify the integrity of model updates across the network, effectively mitigating risks associated with model poisoning and adversarial interference. Our approach, novel in its application of CFA principles to FL, ensures contributions from participating nodes are authentic and untampered, thereby enhancing system resilience without compromising computational efficiency or model performance. Empirical evaluations on benchmark datasets, MNIST and CIFAR-10, demonstrate our framework's effectiveness, achieving a 100\% success rate in integrity verification and authentication and notable resilience against adversarial attacks. These results validate the proposed security enhancements and open avenues for more secure, reliable, and privacy-conscious distributed machine learning solutions. Our work bridges a critical gap between cybersecurity and distributed machine learning, offering a foundation for future advancements in secure FL.

翻译：联邦学习（FL）作为一种分布式机器学习范式的出现，带来了新的网络安全挑战，尤其是威胁模型完整性和参与者隐私的对抗性攻击。本研究提出了一种创新性安全框架，其灵感来源于传统网络安全中用于确保软件执行完整性的控制流证明（CFA）机制。通过将数字签名和密码哈希技术集成到联邦学习框架中，我们能够对网络中模型更新的完整性和真实性进行认证与核验，从而有效缓解与模型中毒和对抗干扰相关的风险。该方法创新性地将CFA原则应用于联邦学习，确保参与节点的贡献真实且未被篡改，进而在不牺牲计算效率或模型性能的前提下，增强了系统韧性。在MNIST和CIFAR-10基准数据集上的实证评估表明，该框架在完整性验证与认证方面实现了100%的成功率，并展现出显著的对抗攻击韧性。这些结果验证了所提安全增强方案的有效性，并为开发更安全、更可靠且更注重隐私保护的分布式机器学习解决方案开辟了新路径。本研究填补了网络安全与分布式机器学习之间的关键空白，为未来联邦学习安全领域的进步奠定了基础。