Encryptability As a Coordinate Choice: Depth-One Homomorphic Federated Learning of Quantum Neural Networks
A new arXiv preprint identifies the coordinate system used to parameterise variational quantum circuit weights as the core obstacle to homomorphically encrypted federated training. Standard representations such as Euler angles or discrete alphabets map SU(2) rotation updates to expressions that are not low-degree, breaking the assumptions required for encrypted aggregation. The work frames encryptability as a coordinate choice for depth-one quantum neural networks.
AI analysis — not reported by the source
What this could mean
- 0–2 yearsPlausible
This could make privacy-preserving federated training of small quantum neural networks practical on near-term simulators and few-qubit devices within two years.
If the proposed coordinate choice indeed converts SU(2) updates into low-degree polynomials, existing homomorphic encryption libraries can be reused without new hardware, and depth-one quantum neural networks are already classically simulable for small qubit counts, so pilot deployments could follow quickly.
This is a brief. The day’s lead story carries the full analysis.