Large language models (LLMs) exhibit abilities beyond natural language modelling and text generation. Recent advances in their reasoning capabilities have spurred interest in applying LLMs to complex scientific tasks requiring deep domain expertise and sophisticated reasoning. Quantum computing, as a highly specialised field with significant knowledge barriers and hardware constraints, could greatly benefit from such advancements. However, a key open question that first must be answered is: How can we develop fine-tuning pipelines that instil genuine quantum reasoning in LLMs, rather than task-specific pattern matching? We study this question through quantum circuit simulation as a training objective, where the model must predict the measurement probability distribution resulting from a sequence of quantum gate operations. We propose and compare two fine-tuning pipelines: (1) Supervised Fine-Tuning (SFT) on explicit gate-by-gate state-vector simulation traces, and (2) a two-stage SFT+Group Relative Policy Optimisation (GRPO) approach that sequentially applies SFT followed by GRPO with verifiable rewards. Our findings show that SFT achieves near-perfect in-distribution and gate-count extrapolation accuracy, significantly outperforming both the base model and the GPT-OSS-120B baseline. SFT+GRPO trades some in-distribution precision for better generalisation to larger qubit systems that SFT alone cannot handle. Both pipelines significantly outperform the baselines, demonstrating that targeted fine-tuning on explicit reasoning traces is an effective strategy for advancing quantum reasoning in LLMs.

Adversarial perturbations threaten machine learning classifiers, including variational quantum classifiers. We show that finite quantum measurement statistics (shot noise) act as a built-in defense against gradient-based test-time attacks whose cost scales unfavorably for the attacker. Because every gradient component must be inferred from repeated circuit executions under any unbiased gradient-estimation rule, white-box extraction consumes a dimension-dependent measurement budget that measurement grouping cannot remove in expressive circuits. Under stated assumptions, single-step attacks need at least quadratically many shots in the input dimension d, growing as d^5/2 under norm-concentration scaling, with a sufficient-budget analysis for iterative attacks via stochastic gradient Langevin dynamics. Simulations up to 784 input dimensions validate the law: the realized total budget is the d^5/2 geometric floor for plateau-mitigated models and grows as d^3.00 for the tested deep circuits, whose gradient norms decay with dimension absent barren-plateau mitigation; folding the measured gradient norm back in recovers the parameter-free d^3/2 shot-noise geometry. Against a matched classical baseline whose attack overhead is dimension-independent (the cheap-gradient principle of automatic differentiation), the quantum gradient cost ratio grows empirically as d^3.00, so the attacker's relative cost diverges as the model scales. Experiments on a 156-qubit IBM processor (ibm_boston, 4-qubit circuits, d=12) reproduce the effect: at matched budgets the device attack tracks the ideal within a few percent, with the high-shot gradient faithful to the exact one. The defense operates precisely when the forward map is classically hard to simulate: only then is a white-box attacker denied the simulate-and-backpropagate shortcut and must pay the measurement cost we quantify.

Rapid advancements in cloud based platforms providing access to quantum computing capabilities have opened up several challenges for efficient usage of these highly delicate and costly devices. Although most of the current systems use a priority based access protocol, they are unable to fully support reliable, efficient, and scalable execution of larger-scale applications. To overcome this limitation, we propose a comprehensive solution for efficient allocation of quantum programs to appropriate quantum devices, considering all the relevant cost metrics into account including, fidelity, execution time and communication overhead. We also formulate use-cases for distributed quantum workflow and propose modified graph based algorithms to solve for allocation of such use-cases, assuming a hybrid classical-quantum network. Since hardware advancements in large standalone devices is an ongoing process, it is critical to investigate such distributed workflows to maximize the best utilization of current NISQ devices. Our empirical study shows that the proposed techniques perform better than state-of-the-art methods for almost all evaluation parameters, with average improvements of approximately 5% in execution time, 30% in communication overhead, 40% in wait time and 2% in fidelity, providing better solutions to efficient allocation strategies.

A central design principle in modern machine learning and artificial intelligence is to align a model's inductive bias with the structure of its input data. For matrix-valued inputs, relevant matrix-level relationships can be characterised through spectral values and spectral subspaces; however, common coordinate-wise rotation-gate data-encoding unitaries used in most quantum machine learning models do not explicitly construct such a matrix-level representation. We introduce Quantum Spectral Models (QSMs), in which we construct the generator of the data-encoding unitary directly from each input matrix. We study three QSM variants based on symmetric, global block, and non-overlapping patch-local block Hamiltonians. Their outputs admit truncated Fourier representations in which input-dependent spectral gaps supply candidate phase carriers, while spectral subspaces help determine their coefficients. We evaluate the QSMs and comparison quantum models on two matrix representations of Pendigits and two controlled synthetic tasks defined by spectral statistics. At the largest evaluated circuit depth, QSM variants lead the tested quantum models in mean test accuracy across all four benchmarks. The patch-local QSM leads on Pendigits, whereas the global block-Hamiltonian QSM leads on the controlled spectral tasks. Ablations show a task-dependent reversal: subspace-preserving controls perform better on Pendigits, whereas spectral-value-only controls lead among the tested ablations on the synthetic tasks. Together, these results shed new light on quantum machine-learning model design by showing how input-conditioned spectral representations can provide an analysable inductive bias, while offering a broader perspective on structure-aware model design in machine learning and artificial intelligence.

Learning with large-scale datasets and information-critical applications, such as in High Energy Physics (HEP), demands highly complex, large-scale models that are both robust and accurate. To tackle this issue and cater to the learning requirements, we envision using a federated learning framework with a quantum-enhanced model. Specifically, we design a hybrid quantum-classical long-shot-term-memory model (QLSTM) for local training at distributed nodes. It combines the representative power of quantum models in understanding complex relationships within the feature space, and an LSTM-based model to learn necessary correlations across data points. Given the computing limitations and unprecedented cost of current stand-alone noisy-intermediate quantum (NISQ) devices, we propose to use a federated learning setup, where the learning load can be distributed to local servers as per design and data availability. We demonstrate the benefits of such a design on a classification task for the Supersymmetry(SUSY) dataset, having 5M rows. Our experiments indicate that the performance of this design is not only better that some of the existing work using variational quantum circuit (VQC) based quantum machine learning (QML) techniques, but is also comparable (Δ∼± 1%) to that of classical deep-learning benchmarks. An important observation from this study is that the designed framework has <300 parameters and only needs 20K data points to give a comparable performance. Which also turns out to be a 100× improvement than the compared baseline models. This shows an improved learning capability of the proposed framework with minimal data and resource requirements, due to the joint model with an LSTM based architecture and a quantum enhanced VQC.

Combinatorial optimization problems are typically formulated using Quadratic Unconstrained Binary Optimization (QUBO), where constraints are enforced through penalty terms that introduce auxiliary variables and rapidly increase Hamiltonian complexity, limiting scalability on near term quantum devices. In this work, we systematically study Quadratic Unconstrained D-ary Optimization (QUDO) as an alternative formulation in which decision variables are encoded directly in higher dimensional Hilbert spaces. We demonstrate that QUDO naturally captures structural constraints across a range of problem classes, including the Traveling Salesman Problem, two variants of the Vehicle Routing Problem, graph coloring, job scheduling, and Max-K-Cut, without the need for extensive penalty constructions. Using a qudit-level implementation of the Quantum Approximate Optimization Algorithm (qudit QAOA), we benchmark these formulations against their binary QUBO counterparts and exact classical solutions. Our study show consistently improved approximation ratios and substantially reduced computational overhead at comparable circuit depths, highlighting QUDO as a scalable and expressive representation for quantum combinatorial optimization.

Reinforcement learning (RL) is one of the most practical ways to learn from real-life use-cases. Motivated from the cognitive methods used by humans makes it a widely acceptable strategy in the field of artificial intelligence. Most of the environments used for RL are often high-dimensional, and traditional RL algorithms becomes computationally expensive and challenging to effectively learn from such systems. Recent advancements in practical demonstration of quantum computing (QC) theories, such as compact encoding, enhanced representation and learning algorithms, random sampling, or the inherent stochastic nature of quantum systems, have opened up new directions to tackle these challenges. Quantum reinforcement learning (QRL) is seeking significant traction over the past few years. However, the current state of quantum hardware is not enough to cater for such high-dimensional environments with complex multi-agent setup. To tackle this issue, we propose a distributed framework for QRL where multiple agents learn independently, distributing the load of joint training from individual machines. Our method works well for environments with disjoint sets of action and observation spaces, but can also be extended to other systems with reasonable approximations. We analyze the proposed method on cooperative-pong environment and our results indicate 10

Artificial Intelligence (AI), with its multiplier effect and wide applications in multiple areas, could potentially be an important application of quantum computing. Since modern AI systems are often built on neural networks, the design of quantum neural networks becomes a key challenge in integrating quantum computing into AI. To provide a more fine-grained characterisation of the impact of quantum components on the performance of neural networks, we propose a framework where classical neural network layers are gradually replaced by quantum layers that have the same type of input and output while keeping the flow of information between layers unchanged, different from most current research in quantum neural network, which favours an end-to-end quantum model. We start with a simple three-layer classical neural network without any normalisation layers or activation functions, and gradually change the classical layers to the corresponding quantum versions. We conduct numerical experiments on image classification datasets such as the MNIST, FashionMNIST and CIFAR-10 datasets to demonstrate the change of performance brought by the systematic introduction of quantum components. Through this framework, our research sheds new light on the design of future quantum neural network models where it could be more favourable to search for methods and frameworks that harness the advantages from both the classical and quantum worlds.

When applying quantum computing to machine learning tasks, one of the first considerations is the design of the quantum machine learning model itself. Conventionally, the design of quantum machine learning algorithms relies on the “quantisation” of classical learning algorithms, such as using quantum linear algebra to implement important subroutines of classical algorithms, if not the entire algorithm, seeking to achieve a quantum advantage through possible run-time accelerations brought by quantum computing. However, recent research has started questioning whether quantum advantage via speedup is the right goal for quantum machine learning (Schuld and Killoran 2022 PRX Quantum 3(3):030101.). Research also has been undertaken to exploit properties that are unique to quantum systems, such as quantum contextuality, to better design quantum machine learning models (Bowles et al. 2023). In this paper, we take an alternative approach by incorporating the heuristics and empirical evidences from the design of classical deep learning algorithms to the design of quantum neural networks. We first construct a model based on the data reuploading circuit (Pérez-Salinas et al. 2020 Quantum 4(226):226) with the quantum Hamiltonian data embedding unitary (Schuld and Petruccione 2021). Through numerical experiments on image datasets, including the famous MNIST and FashionMNIST datasets, we demonstrate that our model outperforms the quantum convolutional neural network (QCNN) (Cong et al. 2019 Nat Phys 15(12):1273–1278) by a large margin (up to over 40

Binary cyclic codes having large dimensions and minimum distances close to the square-root bound are highly valuable in applications where high-rate transmission and robust error correction are both essential. They provide an optimal trade-off between these two factors, making them suitable for demanding communication and storage systems, post-quantum cryptography, radar and sonar systems, wireless sensor networks, and space communications. This paper aims to investigate cyclic codes by an efficient approach introduced by Ding [5] from several known classes of permutation monomials and trinomials over 𝔽_2^m . We present several infinite families of binary cyclic codes of length 2^m-1 with dimensions larger than (2^m-1)/2 . By applying the Hartmann-Tzeng bound, some of the lower bounds on the minimum distances of these cyclic codes are relatively close to the square root bound. Moreover, we obtain a new infinite family of optimal binary cyclic codes with parameters [2^m-1,2^m-2-3m,8] , where m≥ 5 is odd, according to the sphere-packing bound.

Let $p$ be an odd prime and $\mathcal{N}$ be a positive integer such that $p$ is a primitive root modulo $\mathcal{N}$. Define $q=p^{\phi(\mathcal{N})}$. In [5], Cheng and Gao provided a novel method to evaluate the exponential sum given by \begin{equation*} S_{\mathcal{N}}(a, b)=\sum_{x \in \mathbb{F}_{q}^{*}} \chi\left(a x^{\frac{q-1}{\mathcal{N}}}+b x\right) \end{equation*} where $\mathcal{N}=2,4, \ell^{m}$ or $2 \ell^{m}$ with $\ell$ be an odd prime distinct from $p$. For $p=3$ and $q=3^{\phi\left(\ell^{m}\right)}$, they investigated the linear code $\mathcal{C}_{D}$ by choosing the defining set $D=\left\{x \in \mathbb{F}_{q}^{*}: \operatorname{Tr}\left(x^{\frac{q-1}{e^{m}}}\right)=0\right\}$, and constructed a family of two-weight ternary linear codes with completely determined weight distribution. Very recently, Cheng and Sheng [6] considered a more general defining set $D$ for an arbitrary odd prime $p$ with $D=\left\{x \in \mathbb{F}_{q}^{*}: \operatorname{Tr}\left(x^{\frac{q-1}{2 \ell^{m}}}+\beta x\right)=\alpha\right\}$, where $\alpha \in \mathbb{F}_{p}$ and $\beta \in \mathbb{F}_{q}$. They constructed some families of two-weight linear codes and computed their weight distribution when $\alpha \in \mathbb{F}_{p}$ and $\beta=0$. For $\beta \neq 0$, they determined all possible weights of codewords in $\mathcal{C}_{D}$ and demonstrated that it has at most $p+1$ distinct nonzero weights. Motivated by the above works, in this paper, for any odd prime $p$ we investigate the linear code $\mathcal{C}_{D}$ for the defining set $D=\left\{x \in \mathbb{F}_{q}^{*}: \operatorname{Tr}\left(x^{\frac{q-1}{e^{m}}}\right)=u, \operatorname{Tr}(x)=0\right\}$, where $u \in \mathbb{F}_{p}$ and determine all possible weights of codewords in $\mathcal{C}_{D}$, demonstrating that it has at most $p+1$ distinct nonzero weights. Additionally, we determine the minimum distance of $\mathcal{C}_{D}^{\perp}$ and prove that $\mathcal{C}_{D}^{\perp}$ is optimal under a certain condition with respect to the Sphere packing bound.

In this paper, we develop a generalized construction framework of zero-correlation zone (ZCZ) sequence sets based on para-unitary (PU) matrices. We show that any square PU matrix with unimodular sequences can generate ZCZ sequence sets by perfectly controlling its expanded product form, thereby improving upon existing results. The key idea is to impose specific conditions on the expanded product forms of PU matrices to enable accurate computation of periodic correlation functions. In addition, each constructed ZCZ sequence set is a PU column vector and hence it represents a Golay complementary set (GCS). In this work, we call them Golay-ZCZ sequence sets. The lengths, set sizes and ZCZ widths of the constructed Golay-ZCZ sequence sets are all flexible with the form of non-power-of-two. These Golay-ZCZ sequence sets can be directly applied to practical training-based channel estimation schemes in multiple-antenna systems.

By leveraging the principles of quantum mechanics, quantum machine learning (QML) opens doors to novel approaches in machine learning (ML) and offers potential speedup. However, ML models are well documented to be vulnerable to malicious manipulations, and this susceptibility extends to the models of QML. This situation necessitates a thorough understanding of QML's resilience against adversarial attacks, particularly in an era where quantum computing capabilities are expanding. In this regard, this article examines model-independent bounds on adversarial performance for QML. To the best of our knowledge, we introduce the first computation of an approximate lower bound for adversarial error when evaluating model resilience against sophisticated quantum-based adversarial attacks. Experimental results are compared to the computed bound, demonstrating the potential of QML models to achieve high robustness. In the best case, the experimental error is only 10% above the estimated bound, offering evidence of the inherent robustness of quantum models. This work not only advances our theoretical understanding of quantum model resilience but also provides a precise reference bound for the future development of robust QML algorithms.

The Quantum Approximate Optimization Algorithm (QAOA) is a prominent variational algorithm for solving combinatorial optimization problems such as the Max Cut problem. A key challenge in QAOA is the efficient identification of variational parameters (γ, \{beta}) that yield high-quality solutions. In this work, we investigate swarm optimization methods as robust strategies for exploring the QAOA parameter space. We evaluate Particle Swarm Optimization (PSO), Fully Informed Particle Swarm Optimization (FIPSO), Quantum Particle Swarm Optimization (QPSO), and an Adam-assisted FIPSO variant on weighted MaxCut instances across multiple system sizes, circuit depths, and noise regimes, including shot noise. Our results show that these methods achieve lower approximation gaps and more stable convergence compared to standard optimizers such as Adam, COBYLA, and SPSA. In particular, we observe that swarm methods maintain superior performance under noisy and shot limited conditions. These findings suggest that population based search is effective for navigating the complex QAOA landscape and is a promising approach for parameter optimization in near-term quantum algorithms.

Binary plaintext-checking oracle-based chosen-ciphertext attacks have been suggested for retrieving the long-term private key of ML-KEM. Although previous works have shown efficiency, we found significant challenges while mounting attacks in real-world scenarios with substantial environmental noise. In this work, we introduce a Differential Power Analysis-based self-comparison oracle to address the difficulties of creating a near-perfect binary plaintext-checking oracle under such conditions. We validate our approach on a real-world PCB, demonstrating that it simplifies the requirements for executing the chosen-ciphertext attack and offers a more straightforward and effective method for targeting an unknown real-world ML-KEM device.

The relation between stabilizer codes and binary codes provided by Gottesman and Calderbank et al. is a celebrated result, as it allows the lifting of classical codes to quantum codes. An equivalent way to state this result is that the work allows us to lift decoders for classical codes over the Hamming metric to decoders for stabilizer quantum codes. A natural question to consider: Can we do something similar with decoders for classical codes considered over other metrics? i.e., Can we lift decoders for classical codes over other metrics to obtain decoders for stabilizer quantum codes? In our current work, we answer this question in the affirmative by considering classical codes over the symbol-pair metric. In particular, we present a relation between the symplectic weight and the symbol-pair weight and use it to improve the error correction capability of CSS-codes (a well-studied class of stabilizer codes) obtained from cyclic codes.

In frequency hopping (FH) sequence design, the wide gap is an important performance indicator. In this paper, the average Hamming correlation of wide gap FH sequence (WG-FH sequence) sets with low hit zone, which has not yet been reported previously, is studied. A lower bound on the average Hamming auto-correlation and the average Hamming cross-correlation of low-hit-zone FH sequence sets is firstly derived. The new lower bound includes the bound for conventional FH sequence sets derived by Peng et al. as a special case. Then a construction of WG-FH sequence sets with multiple low hit zones is presented, which are optimal by the new bound for these low hit zones. Finally, another class of optimal WG-FH sequence sets with multiple low hit zones is presented, which have larger wide gap.

Machine-learning-assisted side-channel analysis (ML-assisted SCA) automates the procedure of analyzing side-channel activities to reconstruct secrets. Although ML-assisted SCA does produce promising results, it is hard to determine whether its machine-learning model tends to reconstruct secrets or generate new instances. In this paper, we revisit the first general ML-assisted SCA framework for media software (Yuan et al. USENIX Security 2022), which we refer to as the Manifold-SCA framework, with a case study of reconstructing images from cache activities. We show that Manifold-SCA tends to generate images more than reconstruct them. Inspired by the autoencoder implemented in the Manifold-SCA framework, we theoretically and experimentally show that an autoencoder is sufficient to reconstruct images from cache activities. Through three ablation studies, we show that an autoencoder outperforms the Manifold-SCA framework under all scenarios. In the end, we apply an autoencoder to analyze practical cache activities collected by a profiling-based Prime+Probe attack, and show that an autoencoder can reconstruct partial pixel-related activities, but these activities are insufficient to reconstruct images due to the information loss in the activities.

DNA sequences are prone to creating secondary structures by folding back on themselves by non-specific hybridization of its nucleotides. The formation of large stem-length secondary structures makes the sequences chemically inactive towards synthesis and sequencing processes. Furthermore, in DNA computing, other constraints like homopolymer run length also introduce complications. In this paper, our goal is to tackle the problems due to the creation of secondary structures in DNA sequences along with constraints such as not having a large homopolymer run length. This paper presents families of DNA codes with secondary structures of stem length at most two and homopolymer run length at most four. We identified ℤ_11 as an ideal structure to construct DNA codes to avoid the above problems. By mapping the error-correcting codes over ℤ_11 to DNA nucleotides, we obtained DNA codes with rates 0.5765 times the corresponding code rate over ℤ_11 , including some new secondary structure-free and better-performing codes for DNA-based data storage and DNA computing purposes.

DNA sequences are prone to creating secondary structures by folding back on themselves by non-specific hybridization among its nucleotides. The formation of secondary structures makes the sequences chemically inactive towards synthesis and sequencing processes. In this letter, our goal is to tackle the problems due to the creation of secondary structures in DNA sequences along with constraints such as not having a large homopolymer run length. In this paper, we have presented families of DNA codes with secondary structures of stem length at most two and homopolymer run length at most four. By mapping the error correcting codes over $\Z_{11}$ to DNA nucleotides, we obtained DNA codes with rates $0.5765$ times the rate of corresponding code over $\Z_{11}$, which include some new secondary structure free and better-performing codes for DNA based data storage and DNA computing purposes.