Quantum noise is known to strongly affect quantum computation, thus potentially limiting the performance of currently available quantum processing units. Even models of quantum neural networks based on variational quantum algorithms, which were designed to cope with the limitations of state-of-the art noisy hardware capabilities, are affected by noise-induced barren plateaus, arising when the noise level becomes too strong. However, the generalization performances of such quantum machine learning algorithms can also be positively influenced by a proper level of noise, despite its generally detrimental effects. Here, we propose a pre-training procedure to determine the quantum noise level leading to desirable optimization landscape properties. We show that an optimal quantum noise level induces an ``equalization'' of variational parameters: the least important parameters gain relevance in the computation, while the most relevant ones lose it. We analyse this noise-induced equalization through the lens of the Quantum Fisher Information Matrix, thus providing a recipe that allows to estimate the noise level inducing the strongest equalization. Then, we report on extensive numerical simulations providing evidence of the beneficial noise effects in the neighborhood of the best equalization, often leading to improved generalization.
Quantum noise is known to strongly affect quantum computation, thus potentially limiting the performance of currently available quantum processing units. Even models of quantum neural networks based on variational quantum algorithms, which were designed to cope with the limitations of state-of-the art noisy hardware capabilities, are affected by noise-induced barren plateaus, arising when the noise level becomes too strong. However, the generalization performances of such quantum machine learning algorithms can also be positively influenced by a proper level of noise, despite its generally detrimental effects. Here, we propose a pre-training procedure to determine the quantum noise level leading to desirable optimization landscape properties. We show that an optimal quantum noise level induces an ``equalization'' of variational parameters: the least important parameters gain relevance in the computation, while the most relevant ones lose it. We analyse this noise-induced equalization through the lens of the Quantum Fisher Information Matrix, thus providing a recipe that allows to estimate the noise level inducing the strongest equalization. Then, we report on extensive numerical simulations providing evidence of the beneficial noise effects in the neighborhood of the best equalization, often leading to improved generalization.
Variational Quantum Algorithms (VQAs) may be a path to quantum advantage on Noisy Intermediate-Scale Quantum (NISQ) computers. A natural question is whether noise on NISQ devices places fundamental limitations on VQA performance. We rigorously prove a serious limitation for noisy VQAs, in that the noise causes the training landscape to have a barren plateau (i.e., vanishing gradient). Specifically, for the local Pauli noise considered, we prove that the gradient vanishes exponentially in the number of qubits n if the depth of the ansatz grows linearly with n. These noise-induced barren plateaus (NIBPs) are conceptually different from noise-free barren plateaus, which are linked to random parameter initialization. Our result is formulated for a generic ansatz that includes as special cases the Quantum Alternating Operator Ansatz and the Unitary Coupled Cluster Ansatz, among others. For the former, our numerical heuristics demonstrate the NIBP phenomenon for a realistic hardware noise model.
Quantum noise is conventionally viewed as a fundamental obstacle in near-term quantum computing, motivating extensive error correction and mitigation strategies. \REV{We present numerical evidence within an effective noise modeling framework that challenges this consensus. Through experiments on quantum graph neural networks for molecular property prediction, we observe heterogeneous, initialization-dependent responses within this effective noise modeling framework.} Among randomly initialized models with identical architecture, approximately one-third show performance improvement under moderate noise, while a smaller fraction deteriorate and the remainder are marginally affected. We identify a strong negative correlation (r = -0.62) between baseline model performance and noise benefit, suggesting \REV{a regularization-like effect for under-optimized models while disrupting well-converged ones}. The observed optimal noise level falls below theoretical predictions, indicating error cancellation in structured quantum circuits. These findings suggest that, within the adopted effective noise model, noise effects depend critically on initialization quality and need not be uniformly detrimental, motivating structure- and noise-aware optimization strategies.
ABSTRACT In the current quantum computing paradigm, significant focus is placed on the reduction or mitigation of quantum decoherence. When designing new quantum processing units, the general objective is to reduce the amount of noise qubits are subject to, and in algorithm design, a large effort is underway to provide scalable error correction or mitigation techniques. Yet some previous work has indicated that certain classes of quantum algorithms, such as quantum machine learning, may, in fact, be intrinsically robust to or even benefit from the presence of a small amount of noise. Here, we demonstrate that noise levels in quantum hardware can be effectively tuned to enhance the ability of quantum neural networks to generalize data, acting akin to regularisation in classical neural networks. As an example, we consider two regression tasks, where, by tuning the noise level in the circuit, we demonstrated improvement of the validation mean squared error loss. Moreover, we demonstrate the method's effectiveness by numerically simulating QNN training on a realistic model of a noisy superconducting quantum computer.
Without a successful implementation of fault-tolerant quantum error correction, calculations on quantum computers are subject to noise that limits their capabilities. Here, motivated by realistic near-term hardware considerations, we study the impact of uncorrected local noise on logical quantum circuits. We first show that, in the task of estimating observable expectation values, any noise effectively truncates most quantum circuits to logarithmic depth. We then prove that quantum circuits under any non-unital noise do not exhibit barren plateaus for cost functions composed of local observables. However, by using the effective shallowness, we also design an efficient classical algorithm to estimate observable expectation values within any constant additive accuracy, with high probability over the choice of the circuit, in any circuit architecture. Taken together, our results establish that, unless we carefully engineer quantum circuits to take advantage of the noise, noisy quantum circuits are unlikely to offer an advantage over shallow ones for algorithms that output observable expectation value estimates, such as many variational quantum machine learning proposals.
Context: The main obstacle of quantum computing on its way to scalability is the noisy nature of current hardware. Various types of errors lead to incorrect computational results,and measures such as quantum error mitigation are needed to counteract these errors. Many error mitigation techniques require a thorough understanding of the noise and its impact on computations. Therefore, accurate noise models are indispensable to facilitate meaningful quantum computing in the near future. They also enable realistic simulations of quantum devices. Since these devices are currently limited and quantum computing is expensive, researchers typically rely on simulations for testing their algorithms.Objective: In this thesis, we aim to construct a realistic noise model for quantum computing and optimise its model parameters. The model construction should build on the underlying physical processes, and the number of parameters should scale well with the system size. We also address the evaluation and comparison of different noise models by benchmarks using quantum circuits specific to real-world applications. Finally, we aim to develop useful graphical representations of noise models based on the existing quantum circuit model.Method: We present a benchmarking framework for quantum computing noise models and evaluate it based on benchmarking quality attributes from the literature. Moreover, we construct a quantum computing noise model and develop a training procedure for its parameters. We analyse the parameters of the trained noise model and the impact ofdifferent types of errors on computations. We benchmark the noise model with the above approach and compare it to a noise model provided by IBM’s software development kit Qiskit. Finally, we develop an extension of the quantum circuit model to represent noise channels.Result: The benchmarks performed in this thesis show that our noise model predicts noisy hardware behaviour of IBM’s ibmq_manila quantum device equally well as the Qiskit model, if not better, based on quantum circuits commonly used for variational quantum algorithms. Within the trained noise model, readout error has the most detrimental impact on computations. Our benchmarking approach satisfies relevant quality criteria by choosing suitable quantum circuits and objective functions to compare model predictionsto hardware data. Moreover, the quantum circuit model can be extended meaningfully by noise channels.Conclusion: Our benchmarking approach is suitable for evaluating and comparing quantum computing noise models. In future work, volumetric benchmarks should be performed for various noise models and application contexts. The parameter optimisation based on learning with training circuits yields accurate noise models and shows the importance of mitigating measurement error. Different optimisation algorithms and quantum circuits should be explored, and more types of noise should be incorporated into the noise model.
Quantum Machine Learning (QML) offers a paradigm shift in machine learning, potentially surpassing classical approaches due to quantum phenomena. A critical challenge in many QML applications is the need to effectively manage noise within the model parameters, particularly in training datasets with high dimensionality. This paper explores the application of a novel noise matrix optimization algorithm, specifically tailored for QML, leveraging quantum properties to enhance model accuracy and stability. We propose a method that utilizes quantum superposition and entanglement to explore the parameter space more efficiently, leading to improved generalization performance. The core mechanism focuses on manipulating the noise matrix through quantum rotations and entanglement to achieve a more nuanced and robust optimization. We demonstrate the effectiveness of this approach through a series of simulations and preliminary experiments, highlighting its potential to significantly improve model performance in specific scenarios. The paper concludes with a discussion of future research directions and the implications of this quantum optimization technique for the field of machine learning.