FL-QDSNNs: Federated Learning with Quantum Dynamic Spiking Neural Networks
| Autor: | Innan, Nouhaila, Marchisio, Alberto, Shafique, Muhammad |
|---|---|
| Rok vydání: | 2024 |
| Předmět: | |
| Druh dokumentu: | Working Paper |
| Popis: | This paper introduces the Federated Learning-Quantum Dynamic Spiking Neural Networks (FL-QDSNNs) framework, an innovative approach specifically designed to tackle significant challenges in distributed learning systems, such as maintaining high accuracy while ensuring privacy. Central to our framework is a novel dynamic threshold mechanism for activating quantum gates in Quantum Spiking Neural Networks (QSNNs), which mimics classical activation functions while uniquely exploiting quantum operations to enhance computational performance. This mechanism is essential for tackling the typical performance variability across dynamically changing data distributions, a prevalent challenge in conventional QSNNs applications. Validated through extensive testing on datasets including Iris, digits, and breast cancer, our FL-QDSNNs framework has demonstrated superior accuracies-up to 94% on the Iris dataset and markedly outperforms existing Quantum Federated Learning (QFL) approaches. Our results reveal that our FL-QDSNNs framework offers scalability with respect to the number of clients, provides improved learning capabilities, and represents a robust solution to privacy and efficiency limitations posed by emerging quantum hardware and complex QSNNs training protocols. By fundamentally advancing the operational capabilities of QSNNs in real-world distributed environments, this framework can potentially redefine the application landscape of quantum computing in sensitive and critical sectors, ensuring enhanced data security and system performance. Comment: 9 pages, 6 figures |
| Databáze: | arXiv |
| Externí odkaz: |
|