A Federated Unlearning-Based Secure Management Scheme to Enable Automation in Smart Consumer Electronics Facilitated by Digital Twin
Anik Islam, Hadis Karimipour, Thippa Reddy Gadekallu, Yaodong Zhu
Abstract
In consumer electronics, integrating the Internet of Things (IoT) and Artificial Intelligence (AI) has transformed everyday devices into smart, interconnected systems. However, this progress brings significant challenges in resource management, privacy, and security, particularly with the increasing reliance on data-centric technologies like Deep Learning (DL). The introduction of the Right to Be Forgotten (RBF) policy further complicates data management in DL models. This paper presents a new method for automating consumer electronic devices using Federated Learning (FL). This approach involves training devices with the help of a Digital Twin (DT) and securely storing data on a redactable blockchain after each training cycle. An unlearning mechanism in FL is adapted to meet RBF policy requirements, with the redactable blockchain facilitating the necessary data adjustments. Dual authentication methods are used to prevent malicious attacks: a hampel filter and performance checks during training, and a two-phase system comprising an XoR filter and continuous counter checks for request validation. A proof of concept confirms the system’s effectiveness, demonstrating its superior performance compared to existing methods.