Speaker
Description
Quantum key distribution (QKD) in near-term noisy intermediate-scale quantum (NISQ) devices is constrained by hardware noise, limited qubit resources, and dynamic operating conditions, particularly in resource-constrained Internet-of-Things (IoT) networks. Static protocol selection under such conditions can increase quantum bit error rates (QBER) and reduce key generation performance. This work investigates a variational quantum learning framework for noise-aware QKD protocol optimization. We propose a noise-adaptive variational quantum classifier (VQC) that selects among BB84, E91, and GHZ-based protocols by directly minimizing QBER rather than classification accuracy. A scalar depolarizing noise parameter is encoded through uniform single-qubit rotations and processed using a shallow, hardware-efficient variational ansatz with local rotations and ring-connected entangling gates. The circuit output is obtained from a single-qubit Pauli-(Z) expectation value and deterministically mapped to a protocol decision using fixed thresholds. Training employs supervised mean-squared-error minimization with oracle labels generated from protocol-specific QBER simulations. Post-training quantum Fisher information (QFI) analysis is used to assess circuit trainability and parameter sensitivity. Numerical simulations on six-qubit systems under uniform depolarizing noise p ∼ U[0,0.15] yield a mean QBER of 0.0732, representing a 63.2% reduction compared with static protocol selection. The improvement primarily results from suppressing both the mean and variance of QBER rather than regime-dependent adaptive switching. Analysis of the learned policy and QFI spectrum indicates convergence toward a deterministic protocol choice consistent with global risk minimization. These findings demonstrate the potential of variational quantum models for robust QKD protocol optimization in NISQ-constrained quantum communication systems.