Federated Quantum-Inspired Anomaly Detection using Collaborative Neural Clients

Deepthi Godavarthi¹Vigneswara Jaswanth¹Sribidya Mohanty^1,2Polisetty Dinesh¹Rekapalli Venkata Charan Sathvik¹Fernando Moreira^3*

• ¹VIT-AP University, Amaravati, India
• ²Graphic Era Deemed to be University, Dehradun, India
• ³Universidade de Aveiro, Aveiro, Portugal

Despite constant improvement in anomaly detection with the help of deep learning and FL, most existing systems face a number of limitations: (1) Central aggregation of data poses a serious risk to privacy (2) Results are not extendable to heterogeneous and distributed settings and (3) Little consideration has been placed into the quantum-inspired arena of computation and security which can become quite beneficial. Addressing these gaps, this work proposes a quantum-inspired federated learning framework for anomaly detection that guarantees data privacy in a distributed environment. The proposed system follows the client-server approach where multiple clients independently train local feedforward neural network models on private subsets of data, sharing only model parameters with the central server. The server aggregates these updates using the Federated Averaging (FedAvg) scheme to formally build a global model. Currently, the classical deep-learning techniques are being used; however, the framework is designed in a way that allows for the inclusion of quantum machine learning paradigms, hence preparing the system for future developments of speed and security. The proposed methodology yields 79% accuracy, an implementation for a secure, scalable, collaborative system in domains where privacy is of utmost consideration; such areas include cybersecurity, finance, and healthcare. Hence, the work presents not only another practical federated anomaly-detection technique but also grants a foundation for introducing novel quantum methods into secure distributed AI systems.