On the Mitigability of Probabilistic Photonics CNOT Failures of Variational Quantum Classifiers
Muhammad Hosam, Muhammad Ahmed, Norhan Elsayed Amer +2 more
Implementing quantum computers using linear photonics is currently constrained by the low success probabilities of CNOT gates. While the KLM protocol has shown efficient scalability, it still incurs cost in terms of the number of modes and the use of error correcting codes. Variational quantum circuits have shown intrinsic error mitigation properties, however, this generally requires post-processing, essentially requiring extra circuit executions, hurting execution times. Therefore, it is now typical to augment variational circuits with error-correcting codes to mitigate error for all quantum implementation technologies. This paper studies the error mitigation effect of variational quantum classifiers on probabilistic quantum photonics CNOT gates with no extra executions. We have observed that the gate failure noise does not affect all prepared states equally. Moreover, the corresponding error-immune states still provide a high level of expressibility and entangling capability, indicating the possibility for training to achieve error mitigation effects. Moreover, we also observed another subspace of the parameter space in which error can be suppressed through parameter corrections without the need for post-processing. The paper also conducts a photonics simulation study for a range of probabilistic CNOT gates for a simple binary classification problem and studies the effect of circuit hyperparameters on the error mitigation. The simulations show acceptable accuracy results on the learning tasks, showing the possible utility of linear photonics in quantum learning classification tasks.
IEEE International Conference on Quantum Computing and Engineering (QCE)•Aug 30, 2025•1 citation