The interplay between machine learning and quantum computing can lead to unprecedented prospects for both fundamental research and real-world applications. In this new field of quantum machine learning, quantum classifiers, i.e., quantum algorithms for solving classification problems, have attracted the most attention. However, despite the promising performance of these new models for classification tasks, they are complex black-box models. Therefore, like or even more so than classical machine learning models, their use in critical contexts such as medical diagnosis could be hampered without the support of explanations. This paper addresses this challenge by investigating the use of Local Interpretable Model-agnostic Explanations (LIME) for quantum classifiers, with a specific focus on the stability of the explanations. Stability is a key property for ensuring that explanations are consistent and trustworthy, especially when decisions impact sensitive domains. The paper conducts a systematic stability study of LIME in the quantum setting, validating whether a classical explainable AI technique can be meaningfully applied to quantum models. As shown in experiments involving different datasets belonging to critical applications, LIME provides explanations for quantum support vector machines and quantum neural networks that are not only interpretable but also stable. By highlighting the stability of explanations, these findings demonstrate the suitability of LIME as an explainability tool for quantum classifiers.

Quantum backend selection is emerging as a critical step toward achieving quantum computing utility. Indeed, the accuracy of quantum circuit execution strongly depends on the specific quantum machine used, each characterized by a distinct computational noise pattern. Currently, machine learning models for quantum backend selection are being proposed to address this challenge effectively. However, these approaches still lack two fundamental aspects: the integration of execution queue information, which impacts waiting time and scheduling efficiency, and the explainability of the selected backend, which is crucial to understand and trust the decision-making process. This paper introduces the first fuzzy-based machine learning approach for quantum backend selection capable of addressing both aspects. Experimental results show that the proposed method achieves competitive performance compared to state-of-the-art crisp classifiers, while selecting the backend by incorporating execution waiting times and exposing the circuit features underlying the selection decision.

Industrial systems generate large volumes of multivariate time series data that are complex, dynamic, and affected by noise, where early anomaly detection is critical to ensure operational safety and reliability of the system. Conventional machine learning methods often struggle with the non-linear behaviors, temporal dependencies, and subtle or latent faults that characterize real-world industrial environments. This paper proposes a hybrid anomaly detection framework that integrates Timed Automata to model the dynamic evolution of system behaviors with a Quantum Fidelity-based Fuzzy C-Means clustering algorithm to identify anomalous patterns. The proposed approach is validated on the Skoltech Anomaly Benchmark (SKAB), which simulates realistic industrial scenarios by injecting temporally localized anomalies that are sometimes masked by normal process fluctuations, making them particularly challenging to detect. The experiments provide a comparative analysis with state-of-the-art classical methods. The results highlight the potential of combining symbolic temporal modeling with quantum-inspired clustering to enhance anomaly detection, as demonstrated in an example of a complex, dynamic, and noisy industrial environment.

Quantum computing represents a promising frontier for machine learning, offering new tools to tackle complex problems by leveraging high-dimensional feature spaces and the unique properties of quantum systems. In this context, quantum kernels have emerged as a particularly compelling technique, capable of revealing structures in data that are difficult to detect using classical methods (Schuld and Killoran, 2019 [5]). However, their practical impact on real-world applications remains largely unexplored. This work proposes a hybrid quantum-classical architecture for anomaly detection in industrial cyber-physical systems, using the SWaT testbed as a case study. In complex environments like SWaT, sophisticated cyber-attacks can mimic normal system behavior, making them difficult to identify (Goh et al., 2016 [1]). Quantum kernels may help distinguish such anomalies by enabling more expressive data representations. Experimental results demonstrate the viability of hybrid quantum approaches, providing promising indications for future developments in quantum-supported industrial system security.

Quantum Neural Networks (QNNs) are a promising tool for addressing the limitations of their classical counterparts in many applications, including healthcare. However, most existing approaches rely on gradient-based optimization methods, which require a large number of quantum circuit calls and are affected by barren plateaus, limiting their applicability on current Noisy Intermediate-Scale Quantum (NISQ) devices. This work investigates Evolutionary Quantum Neural Networks (EQNNs) for heart disease classification using the Cleveland dataset. The proposed framework replaces gradient-based training with a Genetic Algorithm (GA), enabling parameter optimization without explicit gradient computation and reducing the dependence of training cost on the number of circuit parameters. Experimental results compare the GA with the ADAM optimizer across different circuit sizes and under both ideal and noisy execution conditions using a mock IBM backend. The results demonstrate that EQNNs achieve competitive classification performance while significantly reducing the number of required quantum circuit evaluations, highlighting their potential as a resource-efficient solution for medical diagnosis on NISQ devices.

In the field of quantum computing, Variational Quantum Algorithms (VQAs) combine quantum circuits with classical optimization to perform several tasks including classification and regression. Although gradient-based methods are commonly used for their training, these methods suffer from barren plateaus, noise sensitivity, and high computational costs. In this scenario, Evolutionary Algorithms (EAs) provide a gradient-free alternative for training VQAs enabling the discovery of high-quality solutions while requiring limited hardware resources. Unfortunately, the number of software tools aimed to the application of EAs for training of VQAs is limited and, moreover, they rarely exploits GPU acceleration. In order to bridge this gap, this work presents f-EVOVAQ, a Python package that extends the EVOVAQ framework by implementing vectorized evolutionary operators on GPU using CuPy arrays. The package enables efficient population-level optimization while remaining user-friendly for non-experts in evolutionary computation. Experiments on quantum classification tasks demonstrate significant reductions in optimization time and improved scalability, with limitations arising from GPU–CPU data transfers.

Quantum computing and artificial intelligence are two highly topical fields of research that can benefit from each other's discoveries by opening a completely new scenario in computation, that of quantum artificial intelligence. Indeed, on the one hand, artificial intelligence algorithms can be made computationally more efficient due to the potential speedup enabled by quantum phenomena; on the other hand, the complex development of quantum computing technologies and methodologies can be properly supported by the use of classical artificial intelligence approaches. The "entanglement" of these two disciplines is opening up completely new directions in computer science research, and this survey aims to provide a systematic and taxonomic overview of the work that has already been done and that which will begin in the near future.

Quantum Genetic Algorithms (QGAs) are emerging as a new class of algorithms that exploit the peculiarities of quantum computers to outperform classical evolutionary approaches in solving complex optimization problems. Despite the theoretical promises, their applicability has been strongly affected by the limitation of current noise-prone quantum hardware. It has only recently been demonstrated that acceptable levels of quantum noise can enhance the exploration capability of QGA in constrained optimization problems. The research corroborates the initial findings, providing further evidence of the efficacy of these algorithms by successfully resolving the NP-hard Ising Problem through the utilization of a QGA equipped with a quantum mating operator. The experimental results show that noisy simulations of QGA are capable of identifying better solutions than genetic algorithms that are equipped with state-of-the-art classical mating operators.

Recently, the quantum computer's ability to perform reliable computations beyond the capabilities of classical computing methods is referred to as quantum utility. To achieve this ability, it is becoming increasingly important to assess quantum algorithm performance in practical applications. Starting from this consideration, this work investigates the performance of the well-known Quantum Approximate Optimisation Algorithm (QAOA) with a Genetic Algorithm-based training in solving a real-world problem in the domain of power systems. As shown in the reported experiments using both an ideal simulator and a real IBM quantum processor, QAOA empowered by genetic algorithms outperforms the compared algorithms, particularly on real quantum hardware at the highest QAOA circuit depth.

The optimization of the limited resources of current quantum devices hinges on the implementation of quantum-compiling techniques. The process of transforming high-level quantum algorithms into executable quantum circuits has a significant impact on the performance and fidelity of computations. One of the most critical aspects of the compiling stack is the synthesis of quantum circuits, which aims to minimize the depth and number of gates, as well as the error rate. Nevertheless, the optimal synthesis algorithms are extremely computationally expensive. This is why synthesis algorithms based on reinforcement learning (RL) have recently been introduced. However, such algorithms are constrained by the topology of quantum processors, necessitating retraining of the RL agent for different topologies. To achieve a balance between adaptability, optimality, and computational cost in RL-based quantum circuit synthesis techniques, this work proposes an RL-based quantum circuit synthesis algorithm of Clifford operators that addresses different topologies simultaneously. Experimental tests demonstrate that our synthesizer outperforms state-of-the-art greedy techniques while being significantly faster than boolean satisfiability solvers (SAT).

Currently, quantum processors belong to the so-called Noisy Intermediate-Scale Quantum (NISQ) era due to the fact that they are prone to errors and characterized by a limited number of qubits. In this scenario, solving discrete optimization problems requires the use of efficient encodings capable of limiting the use of resources of NISQ devices. For this reason, a new encoding scheme has recently been proposed where discrete classical variables are encoded in non-orthogonal states of a quantum system. This idea, integrated with Variational Quantum Eigensolvers (VQEs), has shown good performance in solving complex optimization problems by significantly reducing the number of qubits used. VQEs involve parameterized circuits that are typically trained by means of gradient-based optimization techniques. However, these techniques can suffer from several issues including the barren plateau problem. In order to overcome these issues, this paper introduces, for the first time, the use of a gradient-free technique such as Genetic Algorithms (GAs) to optimize the parameters of VQEs combined with this new encoding based on non-orthogonal states. As shown in the experimental session, GAs outperform the other state-of-the-art gradient-free optimizers in solving the well-known Max k-Cut problem.

Hand gesture recognition is an essential task in the developing of upper limb prosthesis control systems. This task is typically addressed by classifying surface electromyography (sEMG) signals, i.e. electrical signals generated by muscle activity. Unfortunately, hand gesture recognition results in a very challenging task for several reasons such as signal variability, electrical noise, and the high-dimensionality of sEMG signals. In particular, the high-dimensionality of data makes machine learning models computationally expensive. In this scenario, quantum-powered machine learning models such as the Quantum-enhanced Support Vector Machine (QSVM) may help thanks to their inherently ability to work with large spaces due to the principle of quantum superposition. Starting from this consideration, this paper presents, for the first time, the application of QSVM to classify sEMG signals so as to predict hand gestures. As shown by results, QSVM is a suitable approach to face the hand gesture recognition task.

Mamdani Interval Type-2 Fuzzy Systems (IT2FSs) are widely recognized for their ability to handle uncertainty and imprecision in various domains. However, the computational complexity associated with their rule evaluation and type-reduction processes poses a significant challenge, especially in large-scale systems. This paper introduces a hybrid quantum-classical Interval Type-2 Fuzzy System (QIT2FS) that uses the Quantum Fuzzy Inference Engine (QFIE) to accelerate the inference process. Using the quantum paradigm, QIT2FS achieves exponential speedup in rule evaluation compared to classical IT2FSs, while maintaining equivalent performance in terms of inference accuracy. The proposed framework integrates quantum-based computation for generating type-2 fuzzy outputs, which are subsequently processed using classical type reduction and defuzzification techniques. Experimental evaluations demonstrate the viability of QIT2FS and highlight its potential to address computational bottlenecks in IT2FSs, paving the way for more efficient fuzzy logic-based systems.

Quantum Approximate Optimization Algorithm (QAOA) is a leading variational quantum algorithm for solving combinatorial optimization problems on near-term quantum devices. However, its performance is strongly affected by quantum noise, particularly readout errors. While full readout quantum error mitigation (QEM) is resource-intensive for large qubit systems, we adopt a scalable local approach using the tensor product of single-qubit calibration matrices. In this work, we analyze the local QEM effect on QAOA trained via genetic algorithms (GA-QAOA), solving Max-Cut instances on up to 15 qubits using a mock IBM quantum backend. We compare three scenarios: non-mitigation baseline, post-training mitigation, and step-wise mitigation. Results show that step-wise application improves both optimization performance and solution quality, confirming the benefit of integrating local QEM into the training process for practical use on noisy near-term devices.

In the Noisy Intermediate-Scale Quantum (NISQ) era, efficient quantum circuit synthesis is essential for optimizing limited qubit resources and mitigating noise effects. Quantum Linear Reversible Circuits (QLRCs) are a fundamental class of circuits with applications in quantum compilation or quantum error correction. They are composed exclusively of CX and SWAP gates, which are among the noisiest operations on current quantum hardware. Therefore, their efficient synthesis is pivotal for executing QLRCs on NISQ processors, which have restricted connectivity between qubits. In this direction, we introduce a Reinforcement Learning (RL)-based synthesis approach using Proximal Policy Optimization (PPO) to synthesize QLRCs on generic topologies of up to five qubits. This work extends previous work that did not handle different topologies of quantum processors. Our method dynamically adapts to hardware constraints, learning optimal or near-optimal gate decompositions. We compare our approach against Qiskit’s synthesis methods for QLRCs, demonstrating that our RL-based synthesizer consistently achieves lower CX gate depth. These results highlight the potential of RL-driven quantum circuit synthesis as a powerful alternative to traditional heuristic-based techniques, paving the way for more efficient AI-based quantum compilation strategies in the NISQ era.

Quantum computing is a promising and emerging field of computer science with the potential to improve artificial intelligence techniques, particularly those related to optimization. However, current quantum devices operate in the Noisy Intermediate-Scale Quantum era, in which quantum noise can significantly compromise computational reliability. Evolutionary algorithms are inherently stochastic and may therefore be less sensitive to such noise, or even benefit from it. This study evaluates the performance of a hybrid quantum genetic algorithm on a set of combinatorial and continuous optimization problems under various realistic noise models. The findings demonstrate that quantum noise does not hinder the algorithm's performance and can even enhance its exploratory capabilities in some cases. This suggests that noise-aware quantum evolutionary strategies have the potential to be effective even with today's imperfect quantum hardware.

Quantum operator synthesis is a key task in quantum transpilers, enabling the translation of high-level algorithms into hardware-executable gate sequences by decomposing arbitrary operators into native gates while optimizing for efficiency and constraints. Recently, reinforcement learning (RL) has emerged as a promising approach for quantum circuit synthesis, particularly for Clifford operators. However, the crisp-reward functions used in current RL frameworks may fail to capture the subtleties of complex quantum circuit synthesis tasks, leading to suboptimal designs. In this paper, this gap is addressed by integrating a fuzzy reward into an RL algorithm to synthesize Clifford operators. The fuzzy reward is computed by a fuzzy rule-based system, designed from the analysis of the error rate in current IBM quantum devices. The experiments show that our fuzzy reward accelerates convergence, achieving the same performance in terms of depth and controlled-X (CX) depth as the crisp-reward algorithm, but with fewer training steps. This highlights the efficiency of fuzzy rewards and their potential to improve the synthesis capabilities of RL systems.

Classification of bloodstain patterns at the crime scene is a crucial task in the context of forensic investigations. Traditionally, this task is carried out by forensic experts by applying, unfortunately, time-consuming procedures and subjective judgments. For this reason, artificial intelligence techniques have recently been applied to make this task faster and less subjective. This work presents, for the first time, the application of different deep learning techniques for facing the classification of bloodstain patterns as a multi-class problem. As shown in the experimental session, where a dataset created from scratch has been used, deep learning show optimal performance in facing multi-class classification of bloodstains at the crime scene.