Search papers, labs, and topics across Lattice.
This paper investigates the resource overhead of using Fully Homomorphic Encryption (FHE) with the CKKS scheme in Quantum Federated Learning (QFL) for training Quantum Convolutional Neural Networks (QCNNs). They implemented and evaluated QFL with CKKS-encrypted parameters for training QCNNs on MRI scans to predict brain tumors, marking the first implementation of this specific combination. The results show that the memory and communication overhead introduced by FHE are substantial, creating a trade-off between privacy and model performance due to the need to reduce model parameters.
Securing quantum federated learning with homomorphic encryption comes at a steep cost: memory and communication overheads make deployment challenging, forcing a trade-off with model accuracy.
Quantum Federated Learning (QFL) enables distributed training of Quantum Machine Learning (QML) models by sharing model gradients instead of raw data. However, these gradients can still expose sensitive user information. To enhance privacy, homomorphic encryption of parameters has been proposed as a solution in QFL and related frameworks. In this work, we evaluate the overhead introduced by Fully Homomorphic Encryption (FHE) in QFL setups and assess its feasibility for real-world applications. We implemented various QML models including a Quantum Convolutional Neural Network (QCNN) trained in a federated environment with parameters encrypted using the CKKS scheme. This work marks the first QCNN trained in a federated setting with CKKS-encrypted parameters. Models of varying architectures were trained to predict brain tumors from MRI scans. The experiments reveal that memory and communication overhead remain substantial, making FHE challenging to deploy. Minimizing overhead requires reducing the number of model parameters, which, however, leads to a decline in classification performance, introducing a trade-off between privacy and model complexity.
