Abstract Quantum photonic neural networks are variational photonic circuits that can be trained to implement high‐fidelity quantum operations. However, work‐to‐date has assumed idealized components, including a perfect π Kerr nonlinearity. This work investigates the limitations of non‐ideal quantum photonic neural networks that suffer from fabrication imperfections leading to unbalanced photon loss and imperfect routing, and weak nonlinearities, showing that they can learn to overcome most of these errors. Using the example of a Bell‐state analyzer, the results demonstrate that there is an optimal network size, which balances imperfections versus the ability to compensate for lacking nonlinearities. With a sub‐optimal effective Kerr nonlinearity, it is shown that a network fabricated with current state‐of‐the‐art processes can achieve an unconditional fidelity of 0.905 that increases to 0.999999 if it is possible to precondition success on the detection of a photon in each logical photonic qubit. These results provide a guide to the construction of viable, brain‐inspired quantum photonic devices for emerging quantum technologies.
Abstract Quantum photonic neural networks are variational photonic circuits that can be trained to implement high‐fidelity quantum operations. However, work‐to‐date has assumed idealized components, including a perfect π Kerr nonlinearity. This work investigates the limitations of non‐ideal quantum photonic neural networks that suffer from fabrication imperfections leading to unbalanced photon loss and imperfect routing, and weak nonlinearities, showing that they can learn to overcome most of these errors. Using the example of a Bell‐state analyzer, the results demonstrate that there is an optimal network size, which balances imperfections versus the ability to compensate for lacking nonlinearities. With a sub‐optimal effective Kerr nonlinearity, it is shown that a network fabricated with current state‐of‐the‐art processes can achieve an unconditional fidelity of 0.905 that increases to 0.999999 if it is possible to precondition success on the detection of a photon in each logical photonic qubit. These results provide a guide to the construction of viable, brain‐inspired quantum photonic devices for emerging quantum technologies.
Reconfigurable photonic integrated circuits (PICs) for quantum optical information processing are now routinely made in foundries with hundreds or thousands of components. Such PICs rely on linear optics, meaning that all photon-photon interactions are probabilistic, and hence that complex protocols that, for example, produce entanglement or graph states do so at low rates. Adding optical nonlinearities to the circuits provides an activation function analogous to that of classical neural networks, enabling deterministic photon-photon protocols in ideal circuits. Here, we discuss the next step: quantum photonic neural networks made from more realistic – i.e. imperfect – components. We show that these are capable of near-deterministic entanglement generation, when imperfections such as losses, imperfect routing and sub-optimal nonlinearities are balanced against network size. These results provide a design for real quantum photonic neural networks and demonstrate that they may play important roles in emerging quantum technologies.
We introduce the architecture and timing algorithm to realize a time-bin-encoded quantum photonic neural network (QPNN): a reconfigurable nonlinear photonic circuit inspired by the brain and trained to process quantum information. Unlike the typical spatially-encoded QPNN, time-encoded networks require the same number of photonic elements (e.g. phase shifters or switches) regardless of their size or depth. Here, we present a model of such a network and show how to include imperfections such as losses, routing errors and most notably distinguishable photons. As an example, we train the QPNN to realize a controlled-NOT gate, based on a hypothetical ideal Kerr nonlinearity. We then extend our model to a realistic two-photon nonlinearity due to scattering from a single, semiconductor quantum dot coupled to a photonic waveguide. We show that, using this realistic nonlinearity, the QPNN can be trained to act as a Bell-state analyzer which operates with a fidelity of 0.96 and at a rate only limited by losses. We further show that time gating can raise this fidelity to over 0.99, while still maintaining an efficiency exceeding 0.9. Overall, this work lays a framework for the first QPNN encoded in time, and provides a clear path to the scaling of these networks.
We report on a gate-based variational quantum classifier implemented with single photons and probabilistic gates, to emulate the standard quantum circuit model framework. We evaluate the expressive power of two deployable quantum neural networks (QNNs) by computing their effective dimension, a capacity measure grounded in a proven generalization-error bound, and compare them with classical artificial neural networks (ANNs) of equivalent trainable-parameter count. Supervised binary classification tasks are used to benchmark performance across photonic and superconducting QNNs, both of which exhibit superior converged (lower) cross-entropy loss and (higher) prediction accuracy relative to matched-parameter ANNs. For a nonlinearly separable task, our photonic QNN with a single pair of trainable parameters successfully converged (loss 0.04 and accuracy 100%), whereas the equivalent ANN failed to learn the decision boundary, saturating at random-guessing performance. We simulate photonic quantum circuits, training them on the XOR problem and a two-class Iris subset using gradient-free optimization, and assess their robustness to sampling errors under realistic noise processes including photon loss and phase-shifter imperfections. Circuits with comparatively high effective dimension were deployed remotely on a six-qubit photonic quantum processor, achieving classification accuracies of up to 100% in both online and offline learning settings. Notably, even the simplest QNN deployed, with just two trainable parameters, successfully solved tasks that classically require ANNs with at least quadruple the number of parameters, suggesting an emergent algorithmic advantage. Overall, these results demonstrate a clear proof-of-principle that gate-based QNNs can be realized and trained effectively on current photonic hardware, providing proof of algorithmic advantage on a gate-based photonic QNN.
Advances in quantum technologies are accelerating the demand for optical quantum state sensors that combine high precision, versatility, and scalability within a unified hardware platform. Quantum reservoir computing offers a powerful route toward this goal by exploiting the nonlinear dynamics of quantum systems to process and interpret quantum information efficiently. Photonic neural networks are particularly well suited for such implementations, owing to their intrinsic sensitivity to photon-encoded quantum information. However, the practical realisation of photonic quantum reservoirs remains constrained by the inherently weak optical nonlinearities of available materials and the technological challenges of fabricating densely coupled quantum networks. To address these limitations, we introduce a hybrid quantum-classical detection protocol that integrates the advantages of quantum reservoirs with the adaptive learning capabilities of analogue neural networks. This synergistic architecture substantially enhances information-extraction accuracy and robustness, enabling low-cost performance improvements of quantum light sensors. Based on the proposed approach, we achieved significant improvements in quantum state classification, tomography, and feature regression, even for reservoirs with a relatively small nonlinearity-to-losses ratio U/γ≈0.02 in a network of only five nodes. By reducing reliance on material nonlinearity and reservoir size, the proposed approach facilitates the practical deployment of high-fidelity photonic quantum sensors on existing integrated platforms, paving the way toward chip-scale quantum processors and photonic sensing technologies.
Neuromorphic (brain-inspired) photonics accelerates AI 1 with high-speed, energy-efficient solutions for RF communication 2 , image processing 3 , 4 , and fast matrix multiplication 5 , 6 . However, integrated neuromorphic photonic hardware faces size constraints that limit network complexity. Recent advances in photonic quantum hardware 7 and performant trainable quantum circuits 8 offer a path to more scalable photonic neural networks. Here, we show that a combination of classical network layers with trainable continuous variable quantum circuits yields hybrid networks with improved trainability and accuracy. On a classification task, these hybrid networks match the performance of classical networks nearly twice their size. These performance benefits remain even when evaluated at state-of-the-art bit precisions for classical and quantum hardware. Finally, we outline available hardware and a roadmap to hybrid architectures. These hybrid quantum-classical networks demonstrate a unique route to enhance the computational capacity of integrated photonic neural networks without increasing the network size.
Quantum machine learning represents a promising approach to disclose novel possibilities for quantum information protocols. Thus, recent efforts have been dedicated to determine how to merge, or extend, machine learning approaches in the quantum domain. A relevant example is provided by a quantum-optical neural network (QONN), a class of physical systems that aim to represent the quantum-optical version of classical neural networks. This architecture was first introduced to identify the structure of optical circuits in analogy to their classical counterpart, showing the possibility of tackling different quantum tasks. However, the full potential of such an approach still needs to be explored. In this work, we provide a step forward in this direction by employing it to tackle the fundamental task of optical quantum cloning. The employed architecture has been successfully trained to perform deterministic universal quantum cloning on a photonic platform, achieving results that are very close to that of an optimal universal quantum cloner. Furthermore, we have verified numerically the robustness of this approach with respect to experimental imperfections relevant for future implementations. This work thus shows the capabilities and flexibility of QONNs, which might prove important for designing even more complex optical circuits.