Quantum federated learning with pole-angle quantum local training and trainable measurement.

Area of Science:

• Quantum Computing
• Machine Learning
• Wireless Communication

Background:

• Quantum federated learning (QFL) leverages quantum neural networks (QNNs) for enhanced capabilities.
• Classical QFL frameworks face limitations in flexibility and reliability, especially under resource constraints and varying channel conditions.

Purpose of the Study:

• To propose a novel slimmable QFL (SlimQFL) framework that incorporates QNN-grounded slimmable neural network (QSNN) architectures.
• To enhance the efficiency, flexibility, and reliability of QFL by addressing time-varying wireless channels and computing resource constraints.

Main Methods:

• Development of a SlimQFL framework integrating QSNN architectures.
• Implementation of trainable measurement within the QNN for enhanced QFL.
• Design of QSNN based on separated training and dynamic exploitation of joint angle and pole parameters.

Main Results:

• The proposed SlimQFL framework achieves higher classification accuracy compared to standard QFL.
• The QSNN-based SlimQFL demonstrates improved transmission stability, particularly in poor channel conditions.
• The framework ensures higher efficiency through parameter reduction without performance degradation.

Conclusions:

• The QSNN-based SlimQFL offers a more efficient and robust solution for federated learning in quantum settings.
• The novel design, including trainable measurement and parameter exploitation, significantly improves performance under adverse conditions.
• SlimQFL presents a promising advancement for practical QFL applications in resource-constrained and dynamic environments.