Poster Sessions

Fault-tolerant quantum computiers (FTQCs) relies on fast and efficient quantum error correction. Quantum low-density parity-check (qLDPC) codes provide highly efficient logical qubit encoding, but their practical implementation requires real-time decoders that operate within the coherence time of physical qubits. Here, we formulate qLDPC decoding as a combinatorial optimization problem for annealing machines. We investigate this approach through decoding simulations using simulated annealing (SA) and simulated quantum annealing (SQA), and benchmark its performance against the standard belief propagation (BP) method. While BP provides the shortest absolute decoding time, all methods show similar scaling with system size. Moreover, the annealing-based approach achieves a higher logical error-rate threshold than BP. These results indicate the potential of annealing machines as real-time decoders for qLDPC codes.

Molecular docking pose search is a core task in drug discovery, but its combinatorial complexity makes it computationally demanding. We formulate docking as a quadratic unconstrained binary optimization (QUBO) problem that matches ligand atoms to discretized pocket features and solve it with quantum annealing. Because solution quality on a QPU depends strongly on hardware-facing parameters, including penalties, minor embedding, chain strength, and annealing controls, we introduce a feature atom matching (FAM-QUBO) formulation together with a hardware-aware tuning workflow. To reduce the cost of exploring the penalty landscape on QPU hardware, we use Bayesian optimization for sample-efficient penalty search. We then tune chain strength using embedding-aware rules and examine the effect of annealing schedules, including pause-and-quench, after fixing the QUBO and physical parameters. We validate the sequential tuning framework on two QPU-embeddable targets, 3NQ9 and 4JSZ, from the RCSB Protein Data Bank. Results show improved geometric fidelity of sampled docking poses and better time-to-solution (TTS). Under matched evaluation budgets, Bayesian optimization also identifies effective penalty settings with fewer QPU evaluations than random search. These results support a practical and reproducible workflow for quantum annealing-based molecular docking.

Black-box optimization (BBO) is the problem of searching for an input that minimizes the output of an objective function whose input–output relationship is unknown in an explicit analytical form. As a promising approach to BBO, factorization machine with quadratic-optimization annealing (FMQA) has been proposed [1, 2]. FMQA constructs a surrogate model using a machine learning model called a factorization machine (FM) and generates candidate solutions by optimizing this surrogate model with an Ising machine. The obtained solutions are evaluated using the objective function and added to the training dataset of the FM. By repeating this iterative process, the optimization performance of FMQA is progressively improved.

Since the Ising machine can handle only binary variables, integer and continuous variables must be transformed into binary variables through integer–binary encoding. Several encoding methods have been proposed, such as one-hot encoding and domain-wall encoding. Previous studies have shown that the choice of integer–binary encoding affects both the accuracy of the solution search performed by the Ising machine [3] and the overall optimization performance of FMQA [4]. In the iterative process of FMQA, the encoding method influences two main stages: the surrogate model construction stage and the solution search stage. However, a single encoding method does not necessarily yield the best performance in both stages simultaneously.

Motivated by these observations, this study proposes an FMQA framework that employs different integer–binary encoding methods in the machine learning and solution search stages. We derived a method for transforming between QUBO matrices represented using two encoding methods: one-hot encoding and domain-wall encoding. This method enables the use of different encoding methods in the two stages. When applied to the Rastrigin function, a widely used benchmark in BBO, we demonstrate that our method achieves superior optimization performance compared with conventional methods, and the advantage becomes more pronounced in higher-dimensional problems.
This work is conducted in collaboration with Mayumi Nakano, Yuya Seki, and Shu Tanaka.
[1] K. Kitai et al., Phy. Rev. Res. 2, 013319 (2020).

[2] R. Tamura et al., arXiv:2507.18003 (2025).

[3] J. Chen et al., IEEE Trans. Quantum. Eng. 2, 1 (2021).
[4] Y. Seki et al., arXiv:2209.01016 (2022).

Resonant silencer design involves many variables and costly objective evaluations, making optimization challenging. We compare conventional FMQA (Factorization Machines with Quadratic‑optimization Annealing) against an augmented FMQA that injects engineering‑knowledge‑based dummy data to reduce expensive evaluations. In 50 independent simulations, the augmented method reached the predefined convergence threshold roughly twice as fast as conventional FMQA, demonstrating improved optimization efficiency.

Quantum optimization is a promising approach for solving combinatorial optimization problems. Feedback-based quantum algorithms enable heuristic optimization through quantum Lyapunov control of quantum dynamics [1,2]. However, these methods suffer from

significant computational overhead due to the iterative nature of quantum feedback. Recently, a classical algorithm inspired by feedback-based quantum algorithms, which requires no iterative feedback overhead, has been proposed [3]. While this classical method demonstrates its effectiveness, the performance is limited due to the local nature of its dynamics. Furthermore, the correspondence between quantum and classical algorithms has not been clarified. In this study, we propose classical counterparts of feedback-based quantum algorithms, investigate their properties and clarify the differences between quantum and classical approaches. Our results reveal that quantum algorithms exhibit advantages in obtaining high-quality solutions, whereas feedback-based classical algorithms exhibit faster convergence. These findings provide new insights into the roles of quantum and classical feedback mechanisms in optimization.

This work was done in collaboration with Takuya Hatomura.
[1] A. B. Magann, K. M. Rudinger, M. D. Grace, and M. Sarovar, Phys. Rev. Lett. 129, 250502 (2022).

[2] A. B. Magann, K. M. Rudinger, M. D. Grace, and M. Sarovar, Phys. Rev. A 106, 062414 (2022).
[3] T. Hatomura, Phys. Rev. E 112, 055303 (2025).

Simulated annealing provide heuristic solutions to combinatorial optimization problems. In this approach, the cost function of the problem is mapped onto the energy function of a many-body system, with the goal of guiding the system toward its ground state — that is, the optimal solution. During the time evolution of the system, fluctuations are initially set strong and then gradually weakened over time, thereby promoting exploration of the state space while preventing the system from becoming trapped in local energy minima, and ultimately driving the system toward the global energy minimum. Previous studies have extensively debated the differences between thermal and quantum fluctuations in the annealing process. However, a fundamental understanding of both has yet to be established, and it remains unclear how and for which problems quantum annealing achieves an advantage over thermal annealing. Here we investigated the many-body dynamics under thermal and quantum annealing across all possible interaction networks of the ±J model with up to seven Ising spins. As a result, we found that performance differences between thermal and quantum annealing emerge for certain specific interaction network structures. This indicates that thermal and quantum annealing can be distinguished depending on the structure of the energy landscape. Furthermore, examining the behavior of probability fluxes in state space reveals that while thermal and quantum annealing exhibit globally similar behavior, there exist qualitative differences attributable to quantum tunneling. Additionally, investigating the speed limits of order parameters computed from the probability fluxes shows that quantum annealing can surpass the speed limits found in thermal annealing, and that the order parameter evolves at a rate close to the speed limit in almost all cases.

Optical design heavily relies on the experience and skill of the designer, which presents the challenge of variations in optical performance depending on the designer [1]. While automated design includes global optimization using genetic algorithms, it is not fully automated, and the assistance of the designer is indispensable due to the highly non-linear nature of optical simulations.

FMQA (Factorization Machine with Quantum Annealing / Quadratic-optimization Annealing) is an algorithm that uses an Ising machine to achieve high-speed computation of black-box optimizations [2, 3, 4]. Since Kitai et al. demonstrated its effectiveness in applying FMQA to the design of complex thermal functional metamaterials in 2019 [2], this algorithm has been widely applied to the design of photonic crystals [5] and stacked metamaterials [6].

In this study, we conducted a preliminary investigation into the application of FMQA to optical design. We chose the Cooke Triplet, a design used since the 19th century, as our optimization target [7]. The triplet design can simultaneously correct Seidel's five aberrations (spherical aberration, coma aberration, astigmatism, field curvature, and distortion) and chromatic aberration with three lenses, and is a standard yet challenging design problem in optical design.

We defined Seidel's five aberrations as an evaluation function and treated the optical performance calculation as a black-box function. We implemented FMQA using D-Wave's Hybrid Solver Service (HSS) for Constrained Quadratic Models (CQM) [8, 9]. We explored triplet lens configurations using FMQA and attempted to obtain a lens configuration (positive-negative-positive) that is considered superior empirically and optically. At the conference, we will present the formulation of lens parameters for the Ising machine and report whether the theoretically optimal configuration is successfully derived through this approach.

[References]

[1] R. Kingslake and R. B. Johnson, Lens Design Fundamentals, 2nd ed. Academic Press, 2009.

[2] K. Kitai et al., "Designing metamaterials with quantum annealing and factorization machines," Phys. Rev. Res., vol. 2, p. 013319, 2020.

[3] K. Kitai, "Black-Box Optimization with Quantum Annealing," Ph.D. dissertation, Dept. Comput. Biol. Med. Sci., Univ. of Tokyo, Tokyo, Japan, 2021.

[4] R. Tamura et al., "Black-box optimization using factorization and Ising machines," arXiv:2507.18003, 2025.

[5] T. Inoue et al., "Towards optimization of photonic-crystal surface-emitting lasers via quantum annealing," Opt. Express, vol. 30, no. 24, pp. 43503–43512, 2022.

[6] S. Kim et al., "High-performance transparent radiative cooler designed by quantum computing," ACS Energy Lett., vol. 7, pp. 4134–4141, 2022.

[7] H. D. Taylor, "A Simplified Form and Improved Type of Photographic Lens," British Patent No. 22607, 1893.

[8] D-Wave Systems Inc., "D-Wave hybrid solver service: An overview," 2020. [Online]. Available: https://www.dwavequantum.com/media/4bnpi53x/14-1039a-b_d-wave_hybrid_solver_service_an_overview.pdf

[9] D-Wave Systems Inc., "Hybrid solver for constrained quadratic models," 2021. [Online]. Available: https://www.dwavequantum.com/media/rldh2ghw/14-1055a-a_hybrid_solver_for_constrained_quadratic_models.pdf

A combinatorial optimization problem is defined as the problem of determining an assignment of decision variables that minimizes (or maximizes) a prescribed objective function subject to a specified set of constraints. Since the number of candidate solutions grows exponentially with problem size, finding optimal solutions for large-scale instances is generally intractable. To address this challenge, Ising machines, which obtain the best solution by searching for the ground state of the Ising model—a mathematical framework rooted in statistical mechanics—have attracted considerable attention as promising computational architectures capable of solving combinatorial optimization problems efficiently and with high accuracy.

When solving constrained problems on Ising machines, the penalty function is commonly employed, in which constraints are incorporated into the objective function as penalty terms. However, this approach exhibits a strong dependence on the penalty coefficients μ, often requiring extensive computational effort to achieve convergence. In this study, we apply the augmented Lagrangian method, which has been widely used in continuous optimization for its robustness to penalty parameter tuning, to combinatorial optimization on Ising machines, and investigate whether it improves convergence speed compared to the standard penalty method.

Specifically, we adopt the quadratic knapsack problem—a well-known combinatorial optimization problem with inequality constraints—as the benchmark problem and conduct numerical experiments using Fixstars Amplify AE as the solver. We use time-to-solution (TTS) as the performance evaluation metric and analyze the evolution of the objective function values across successive iterations.

The results demonstrate that the use of the augmented Lagrangian method leads to a reduction in the TTS compared to the conventional penalty function, thereby confirming its effectiveness in improving convergence performance for combinatorial optimization problems. With suitably initialized Lagrange multipliers λ, the method converges to the optimal solution significantly faster than the penalty method in the early iterations. These findings suggest that the augmented Lagrangian function is a promising approach for improving convergence and search performance in combinatorial optimization problems as well as in continuous optimization.

In the development of aerodynamic machinery such as turbomachinery, optimization techniques are widely employed to achieve superior performance. Improving aerodynamic efficiency generally requires analysis based on computational fluid dynamics (CFD), which is highly time-consuming. In optimization processes, repeated CFD simulations are necessary, often resulting in long design cycles and making rapid development challenging. To address this issue, we propose an integrated approach that combines quantum-inspired annealing-based optimization algorithm (FMA), expected to enable faster optimization, with fluid analysis techniques. Applied to the design of a turbo-type hydrogen compressor, this method reduced optimization time compared to conventional optimization workflows while achieving high-performance design. This study indicates that the combination of quantum-inspired optimization and CFD has the potential to accelerate future development cycles for turbomachinery.

Network Function Virtualization (NFV) enables flexible service provisioning by transforming traditional hardware-based network functions into software-based Virtual Network Functions (VNFs). In Multi-access Edge Computing (MEC) environments, a Service Function Chain (SFC) is formed by an ordered set of VNFs connected through virtual links. Efficient deployment of SFCs is crucial for ensuring optimal network performance and reliability. This paper addresses the SFC deployment problem in MEC networks, aiming to minimize overall costs and enhance deployment effectiveness. To tackle this problem, we employ Benders decomposition to decompose the original problem into two subproblems: the master problem, which focuses on VNF placement, and the subproblem, which deals with flow routing. For the master problem, we introduce a novel Quantum Annealing-based approach implemented on the D-Wave platform. The solution to the master problem serves as input for the flow routing subproblem, which is solved using the K-shortest path algorithm. Extensive simulation results demonstrate that the proposed approach significantly reduces deployment costs compared to heuristic baseline methods, thereby improving overall deployment efficiency.
Author: Mai Dinh Cong, Seon-Geun Jeong, Won-Joo Hwang

Reference:

C. Pham, N. H. Tran, S. Ren, W. Saad, and C. S. Hong, “Traffic-aware and energy-efficient vnf placement for service chaining: Joint sampling and matching approach,” IEEE Trans. Serv. Comput, vol. 13, no. 1, pp. 172–185, Feb. 2020

Analog quantum computers provide a promising platform for studying many-body quantum dynamics beyond classical simulation. In this work, we investigate a spacetime-localized response [1], an AdS/CFT(anti-de Sitter/conformal field theory) correspondance-inspired phenomenon, using QuEra Computing’s neutral-atom device Aquila [2]. This response can be realized in the ground state of the one-dimensional transverse-field Ising model near the quantum critical point by applying a perturbation, which allows us to study it experimentally in a spin system. The observed response is related to an effective length scale in the dual AdS spacetime.

However, although the near-critical ground state can be prepared, the subsequent dynamics are sensitive to hardware noise, making the response difficult to observe. We analyze the dominant noise sources via numerical simulations and develop methods to improve the visibility of the response. This talk presents both the noise analysis and experimental results obtained on Aquila.
[1] M. Bamba et al., Phys. Rev. D 109, 126003 (2024)

[2] J. Wurtz et al., arXiv:2306.11727 (2023)

Quantum annealing (QA) machines are expected to solve combinatorial optimization problems efficiently. To solve these problems, they must be formulated as an Ising model. However, embedding the formulated model into the hardware of QA machines remains a significant challenge due to the limited connectivity of physical qubits. This process often requires multiple physical qubits to represent a single logical qubit, a structure known as a "chain." The number of physical qubits representing a single logical qubit is referred to as the chain length, and it is well known that a long average chain length degrades solution accuracy. Therefore, reducing the number of logical variables as a preprocessing step is expected to mitigate these hardware-related issues.

In this study, we propose and analyze a variable reduction method that leverages spin-pair correlation functions to improve the performance of QA machines. Our approach focuses on spin-pair correlation functions and achieves variable reduction by merging specific spin pairs. In the preliminary QA machines execute in a short annealing time and thus preprocessing cost is negligible compared to the main optimization run. To evaluate the performance of the proposed method, we analyze the changes in the expected energy values obtained from QA machines.

For our analysis, we used the Sherrington-Kirkpatrick (SK) model of Ising spin glasses, with the longitudinal magnetic field set to zero for all spins. As a preprocessing step, we first sampled low-energy states of the original problems using QA machines and calculated the correlation functions for all spin pairs at the end of annealing. Our preliminary study indicated that merging spin pairs. We found that merging spin pairs whose correlation functions are close to +1 improves performance, which is physically reasonable because strongly correlated spins tend to take the same value in low-energy states. In contrast, merging pairs with negative correlations leads to degradation. Consequently, we set a threshold and merged all spin pairs whose correlation functions exceed this value. We then experimented ten instances and repeated ten times on the machine to compare the average expected energy values.

Our results show that the proposed method successfully reduced the average chain length when embedding the Ising model into QA machines for all problem instances. Furthermore, the expectation value of energy decreased for almost all instances. Even in cases where performance did not improve, raising the threshold (i.e., merging only more strongly correlated pairs) resulted in lower energy expectation values. These results suggest that our variable reduction method is an effective approach for enhancing the performance of QA machines.

・Tadashi Kadowaki and Hidetoshi Nishimori. Physical Review E, Vol. 58, No. 5, p. 5355, 1998.

・David Sherrington and Scott Kirkpatrick., Phys. Rev. Lett., Vol. 35, pp. 1792–1796, Dec 1975.

・Gomez-Tejedor, et al., arXiv preprint arXiv:2504.13376 (2025).

A financial network is defined as a network composed of asset holdings and cross-holdings among institutions, such as countries and companies. The financial network problem aims to determine the equilibrium state of such a system under nonlinear interactions, including defaults [1]. Since solving this problem enables the prediction of loss propagation associated with defaults of institutions, it is expected to simulate financial crises, which are often triggered by defaults, as seen in 2008. However, the financial network problem is known to be an NP-hard problem [2], and the computational complexity increases exponentially depending on the problem size. To address this issue, previous work utilised a quantum annealer, formulating the problem in a Quadratic Unconstrained Binary Optimisation (QUBO) form, including up to second-order polynomial terms [3]. Ising machines such as quantum annealers attract significant attention as a metaheuristic for combinatorial optimisation problems. Whilst the quantum annealer provides a solution, the problem size that could be handled was limited because of the number of available qubits and couplers. In our work, we employed SQBM+ as a Polynomial Unconstrained Binary Optimisation (PUBO) solver of the simulated bifurcation machine (SBM). Although most ordinary Ising machines accept formulations including up to second-order binary polynomials, such as QUBO and Ising models in their solvers, SQBM+ is able to treat polynomials up to fourth-order. Additionally, as SQBM+ is implemented on a digital circuit, it is free from the hardware limitations of quantum annealers, such as the limited number of qubits and couplers. In order to enhance the effectiveness of this PUBO solver, we formulated the financial network problem as PUBO instead of QUBO. Since PUBO can directly represent higher-order interactions, it eliminates the need for auxiliary variables required in the QUBO reduction, thereby reducing the total number of binary variables. As a result, the PUBO solver can provide appropriate solutions to the problem when the problem size is restricted, whilst still increasing the problem size compared to the previous work.

This work was done in collaboration with Tetsuro ABE, Keita TAKAHASHI, Shuta KIKUCHI and Shu TANAKA.
[1] Elliott, M., Golub, B. & Jackson, M. O. (2014). Financial networks and contagion. American Economic Review, 104(10), 3115-3153.

[2] Hemenway, B. & Khanna, S. (2017). Sensitivity and computational complexity in financial networks. Algorithmic Finance, 5(3-4), 95-110.

[3] Ding, Y., Gonzalez-Conde, J., Lamata, L., Martín-Guerrero, J. D., Lizaso, E., Mugel, S. & Sanz, M. (2023). Toward prediction of financial crashes with a D-wave quantum annealer. Entropy, 25(2), 323.

We present a Hamiltonian-aware variational ansatz for lattice protein folding on Noisy Intermediate-Scale Quantum (NISQ) devices. In variational quantum algorithms for protein-folding Hamiltonians, performance is strongly influenced by the ansatz structure: generic hardware-efficient circuits are easy to implement, but often introduce unnecessary parameters and circuit depth that degrade optimization efficiency and increase sensitivity to noise. To address this limitation, we construct the variational circuit directly from the interaction structure of the encoded folding objective. Our method first analyzes the effective pairwise couplings of the Hamiltonian, identifies the strongest qubit-qubit interactions, and then applies trainable single-qubit rotation layers together with fixed CZ entanglers only on these screened strong-interaction pairs.

This design preserves the most relevant correlation channels while keeping the trainable dimension compact and the executable circuit shallow. Through hyperparameter benchmarking, we find that a shallow operating regime already provides the best practical trade-off between convergence quality and computational cost. We then evaluate the proposed ansatz on peptide-folding benchmark instances spanning 6 to 18 qubits and compare it with TwoLocal and DC-QAOA baselines. Across the tested benchmark set, the Hamiltonian-aware ansatz exhibits substantially lower circuit depth, faster convergence on most instances, and the lowest observed energies on several medium-to-large peptide problems. These results indicate that physics-informed ansatz construction is a practical and scalable strategy for peptide-folding VQE on near-term quantum hardware.

1. Alberto Peruzzo, Jarrod McClean, Peter Shadbolt, Man-Hong Yung, Xiao-Qi Zhou, Peter J. Love, Alán Aspuru-Guzik, and

Jeremy L. O’Brien. A variational eigenvalue solver on a photonic quantum processor. Nature Communications, 5(1):4213,

2014.

2. Abhinav Kandala, Antonio Mezzacapo, Kristan Temme, Maika Takita, Markus Brink, Jerry M. Chow, and Jay M. Gambetta.

Hardware-efficient variational quantum eigensolver for small molecules and quantum magnets. Nature, 549(7671):242–246,

2017.

3. Jarrod R. McClean, Sergio Boixo, Vadim N. Smelyanskiy, Ryan Babbush, and Hartmut Neven. Barren plateaus in quantum

neural network training landscapes. Nature Communications, 9(1):4812, 2018.

4. Harper R. Grimsley, Sophia E. Economou, Edwin Barnes, and Nicholas J. Mayhall. An adaptive variational algorithm for

exact molecular simulations on a quantum computer. Nature Communications, 10(1):3007, 2019.

5. Harper R. Grimsley, George S. Barron, Edwin Barnes, Sophia E. Economou, and Nicholas J. Mayhall. Adaptive, problem-

tailored variational quantum eigensolver mitigates rough parameter landscapes and barren plateaus. npj Quantum Information,

9(1):19, 2023.

6. Pranav Chandarana, Narendra N. Hegade, Iraitz Montalban, Enrique Solano, and Xi Chen. Digitized counterdiabatic

quantum algorithm for protein folding. Phys. Rev. Appl., 20:014024, Jul 2023.

Quantum annealing (QA) is a promising metaheuristic for solving combinatorial optimization problems. Solutions are sought by mapping the given problem onto a ground-state search of an Ising model. For constrained optimization problems, the penalty function method is commonly employed. In this approach, the objective function and the constraints are each formulated as Ising models, and constraint terms weighted by penalty coefficients are added to the objective function.

The penalty function method encounters significant difficulties in both parameter tuning and hardware implementation. Regarding parameter tuning, inappropriate penalty coefficients can lead to a mismatch between the ground state and the actual optimal solution. Regarding hardware implementation, it is necessary to embed the graph structure of the Ising model into the physical graph structure of the hardware [1, 2]. This embedding represents each logical spin using multiple physical spins, which increases the effective number of variables. Consequently, this leads to issues such as degraded solution quality and hardware size limitations.

To address these issues, the spin-variable reduction (SVR) method [3], which reduces variables using equality constraints, has been proposed. Although the SVR method eliminates variables by substituting equality constraints into the objective function, new constraint terms are imposed on the remaining variables. Consequently, penalty coefficients still appear in the Hamiltonian, and their appropriate adjustment remains an essential task for finding optimal solutions. Furthermore, the resulting model in the SVR method is often fully connected, still requiring a large number of physical spins for hardware implementation. The impact of penalty coefficient adjustment on solution accuracy within the reduced Hamiltonian and its effectiveness under hardware size constraints have not yet been fully elucidated.

In this study, we investigate the SVR method using the graph bipartitioning problem (GBP), a representative combinatorial optimization problem. We conducted evaluations using both numerical simulations with the quantum many-body library QuTiP and experiments on the D-Wave Advantage2 hardware. Specifically, we analyzed the sensitivity of the ground-state probability to changes in the penalty coefficient and compared the differences in solution accuracy between simulations and real hardware. Through these evaluations, we compare hardware and numerical simulations. This approach enables a discussion of the impact of hardware-specific physical constraints on the performance of each formulation.

This work is conducted in collaboration with Shuta Kikuchi and Shu Tanaka.

[1] V. Choi, Quantum Inf. Process., 7.5 (2008): 193-209.

[2] V. Choi, Quantum Inf. Process., 10.3 (2011): 343-353.

[3] T. Shirai et al., IEEE Trans. Comput., 72.8 (2023): 2151-2164.

Crowd crush accidents can arise when dense pedestrian streams moving in opposite directions converge within confined urban spaces, causing a critical loss of individual mobility and control. To address this issue, this study proposes a Quantum Annealing (QA)-based directional optimization framework for pedestrian networks. The network is modeled as a weighted graph, in which vertices denote intersections and edges denote walkable paths. The objective is to assign directions to paths to reduce flow conflicts, improve travel efficiency, and mitigate localized congestion. To this end, the framework integrates three complementary strategies: (i) a cycle-based approach to preserve global circulation and network connectivity, (ii) an n-hop-based approach to reduce overall travel distance, and (iii) a flow-conservation approach to balance inflow and outflow at each node. The resulting optimization problem is formulated as a quadratic unconstrained binary optimization (QUBO) model and solved on the D-Wave QA platform. Experimental results show that the proposed method reduces congestion, improves flow stability, and yields lower all-pairs shortest-path (APSP) sums while requiring less computation time than the baseline method.
Kadowaki, T., & Nishimori, H. (1998). Quantum annealing in the transverse Ising model.

Kolmogorov, V. (2009). Blossom V: A new implementation of a minimum cost perfect matching algorithm.

Combinatorial optimization problems are mathematical problems that aim to find a combination of decision variables that satisfies given constraints and minimizes or maximizes an objective function. Ising machines[1], including quantum annealing machines, are expected to solve such problems efficiently; however, their solution accuracy is known to degrade for large-scale problem instances[2].

In this study, we employ Large Neighborhood Search (LNS)[3], a metaheuristic framework that iteratively destroys and repairs parts of a solution to explore a larger search neighborhood. A Previous study[4] integrated Ising machines into the LNS framework to enable their application to large-scale problems and obtaining higher-quality solutions than conventional solvers. However, that approach struggles to maintain solution quality across diverse problem settings because the subproblem size is difficult to tune appropriately. Our proposed algorithm enables finer-grained control over subproblem sizes, allowing a systematic search for the most effective. To evaluate our approach, we applied LNS to the Vehicle Routing Problem (VRP). We addressed a problem instance comprising 5 vehicles and 300 locations, and compared its performance with the conventional LNS.

As a result, the proposed LNS improved the objective function value by approximately 20% compared to the conventional LNS. Furthermore, it required only about 30% of the iterations to obtain solutions of the same quality as the conventional LNS.
[1] N. Mohseni et al., Nature Review Physics 4, 363 (2022).

[2] D. Venturelli et al., Phys. Rev. X 5, 031040 (2015).

[3] P. Shaw, Lect. Notes Comput. Sci. 1520, 417 (1998)

[4] M. Yamashita, master’s thesis, Keio University, Graduate School of Science and Technology (2024).

This paper investigates the application of quantum annealing (QA) for optimizing reconfigurable intelligent surface (RIS)-assisted integrated sensing and communication (ISAC) systems with practical low-resolution phase shifts. In particular, we consider a 2-bit RIS architecture, where the discrete phase constraints lead to a challenging combinatorial optimization problem. To jointly enhance communication and sensing performance, we formulate a sum-SNR maximization problem under quality-of-service (QoS) requirements for both users and the sensing target. By exploiting the quadratic structure of the RIS configuration, the problem is expressed in a unified quadratic form. The constraints are incorporated using Lagrange multipliers, resulting in a quadratic unconstrained binary optimization (QUBO) formulation. Based on this transformation, a QA-based approach is proposed to efficiently solve the resulting problem. Simulation results demonstrate that the proposed method achieves competitive performance and effectively handles the optimization problem compared to conventional classical approaches.
Author: Pham Dang Anh Duc, Won Joo Hwang

Reference Paper:

Krikidis, Ioannis, Constantinos Psomas, and Gan Zheng. "1-bit RIS-aided Index Modulation with Quantum Annealing." IEEE Journal of Selected Topics in Signal Processing 

In the steelmaking process, multiple types of scrap are melted to produce the desired steel. To achieve this, it is necessary to adjust the proportions of the various scrap types to meet the specified chemical composition of the steel produced; however, the variation in chemical composition is significant, making feature extraction difficult. In this study, we performed principal component analysis on the past scrap blending records and identify the optimal subset of principal components that maximizes the clustering performance of principal component scores using black-box optimization. This is expected to enable the efficient identification of factors to be considered during operation. By comparing the selected principal components with historical operational data, we confirmed that certain principal components suggest factors critical to the blending.

Deep learning models are used in critical applications, broadly in computer vision and decision making processes, where is fundamental understanding how and why certain outcomes are produced. In this work, we propose an extension of a first attempt approach to interpreting feature maps (FMs) that are extracted from Convolutional Neural Networks (CNNs) in image classification tasks. In particular, we aim to reformulate the selection of the learned representations as a Quadratic Unconstrained Binary Optimization (QUBO) problem due to the exponential scaling of the corresponding approach, which rapidly becomes intractable under exhaustive search strategies. In this way, we enable the use of both classical and quantum-inspired optimization techniques to efficiently explore the combinatorial search space.

Specifically, we investigate the application of Quantum Annealing (QA) and Simulated Annealing (SA) to identify compact subsets of informative FMs that preserve the predictive performance of the original network while enhancing Interpretability. The proposed framework aims to operate on different state-of-the-art models, from multiple CNN architectures to U-Net and Variational Autoencoder, to achieve a good disentanglement between the feature maps that are learned and activated across the whole structure.

The transition to lattice-based post-quantum cryptography (PQC) relies fundamentally on the Learning With Errors (LWE) problem. Although the security of LWE against classical lattice-reduction attacks has been extensively studied, its resistance to rapidly advancing annealing machines has not yet been sufficiently investigated. In this study, we evaluate the security of LWE by formulating the problem as a quadratic unconstrained binary optimization (QUBO) model through a mixed-integer programming (MIP) approach. We simulate decryption attacks using the CPU-based OpenJij and the GPU-based Fixstars Amplify AE, and compare the results with those obtained using an exact MIP solver, PuLP. Our simulations show that the decryption time of annealing-based attacks scales exponentially with problem dimension, indicating that the security of LWE is fundamentally preserved. However, increasing the number of LWE samples reveals a contrasting scaling behavior: whereas exact solvers become more efficient because the search space is more tightly constrained, annealers exhibit lower success rates due to the increased complexity of the objective function. Based on these machine-dependent behaviors, we propose practical LWE parameter guidelines that explicitly take annealing-based attacks into account.

I would like to apply for a poster presentation.

Combinatorial optimization problems are crucial in industry. However, many COPs are NP-hard, causing the search space to grow exponentially with problem size and rendering large-scale instances computationally intractable. Conventional solvers typically treat problems as monolithic entities, leading to significant efficiency degradation as structural complexity increases. To address this issue, we propose a novel search-space decomposition method that leverages the inherent structure of variables to systematically reduce the size of the master problem. We formulate interaction costs between variables and individual variable costs as a constrained maximum cut problem and convert it into a quadratic unconstrained binary optimization formulation using penalty terms. An Ising-model solver is used to rapidly decompose the problem into independent small-scale subproblems, which are subsequently solved in parallel using mathematical optimization solvers. We validated this method on the capacitated vehicle routing problem. Results demonstrate three significant benefits: a substantial enhancement in feasible solution rates, accelerated convergence, achieving in 1 min the accuracy that the naive method required 30 min to reach, and a variable reduction of up to 95.32%. These findings suggest that search-space decomposition is a promising strategy for efficiently solving large-scale combinatorial optimization problems.
https://arxiv.org/abs/2602.23038

Protein side-chain packing selects one rotamer per residue on a fixed backbone to minimize a pairwise-decomposable energy, yielding a block-wise one-hot constrained combinatorial problem that maps naturally to a quadratic unconstrained binary optimization (QUBO) model. In quantum annealing (QA), hard constraints are enforced via penalty terms scaled by a parameter $\lambda$, making penalty selection a first-order design choice: insufficient penalties permit infeasible solutions, while excessive penalties compress objective contrast and interact adversely with minor-embedding artifacts such as chain breaks. We propose and compare three static, instance-dependent penalty scaling heuristics adapted to rotamer energy tables: (i) UBP, a conservative rule based on the global sum of absolute objective coefficients; (ii) EA-MQCP, an energy-aware rule aggregating per-residue and per-interacting-pair maxima; and (iii) MQCP, a compact rule based on global one-body and two-body maxima. We evaluate these rules on D-Wave quantum annealers across four benchmark instances of increasing size, using a dense logarithmic sweep of $\lambda$ to map empirical feasibility-transition zones and objective-energy convergence landscapes. On small and medium instances, all three heuristics place $\lambda$ within or above the empirical transition zone, achieving near-complete feasibility and recovering solutions close to the known ground state. MQCP, being the most compact, positions closest to the transition boundary and best preserves objective resolution among feasible configurations, while UBP provides the widest safety margin at the cost of over-scaling. EA-MQCP offers a practical intermediate. As instance size grows, however, feasibility saturates even under large penalty magnitudes, suggesting that chain-break-induced decoding failures at the hardware level become the dominant bottleneck rather than penalty insufficiency. These findings demonstrate that effective penalty scaling for QA-based side-chain optimization must account jointly for instance energy statistics and hardware-level embedding characteristics, and that beyond a certain problem scale, penalty tuning alone is insufficient without co-consideration of embedding stability.

1. Harder, T. et al. Beyond rotamers: a generative, probabilistic model of side chains in proteins. BMC Bioinforma. 11, 306,

DOI: 10.1186/1471-2105-11-306 (2010).

2. Guidi, G., Di Tucci, L. & Santambrogio, M. D. Profax: A hardware acceleration of a protein folding algorithm. In 2016

IEEE 2nd International Forum on Research and Technologies for Society and Industry Leveraging a better tomorrow

(RTSI), 1–6, DOI: 10.1109/RTSI.2016.7740584 (IEEE, 2016).

3. Alford, R. F. et al. The rosetta all-atom energy function for macromolecular modeling and design. J. Chem. Theory

Comput. 13, 3031–3048, DOI: 10.1021/acs.jctc.7b00125 (2017).

4. Dauparas, J. et al. Robust deep learning–based protein sequence design using proteinmpnn. Science 378, 49–56 (2022).

5. Misiura, M., Shroff, R., Thyer, R. & Kolomeisky, A. B. Dlpacker: Deep learning for prediction of amino acid side chain

conformations in proteins. Proteins: Struct. Funct. Bioinforma. 90, 1278–1290 (2022).

6. Agathangelou, A., Manawadu, D. & Tavernelli, I. Quantum algorithm for protein side-chain optimisation: Comparing

quantum to classical methods (2025). 2507.19383.

7. Jeong, S.-G., Moon, K.-H. & Hwang, W.-J. Hybrid quantum neural networks for efficient protein-ligand binding affinity

prediction. EPJ Quantum Technol. 12, 120, DOI: 10.1140/epjqt/s40507-025-00419-1 (2025).

8. Ayodele, M. Penalty weights in QUBO formulations: Permutation problems. In Evolutionary Computation in Combinato-

rial Optimization, 159–174, DOI: 10.1007/978-3-031-04148-8_11 (Springer International Publishing, 2022).

13/14

9. García, M. D., Ayodele, M. & Moraglio, A. Exact and sequential penalty weights in quadratic unconstrained binary

optimisation with a digital annealer. In Proceedings of the Genetic and Evolutionary Computation Conference Companion,

GECCO ’22, 184–187, DOI: 10.1145/3520304.3528925 (Association for Computing Machinery, New York, NY, USA,

2022).

10. Qin, J.-Q., Yu, Y. & Wang, X.-B. Variational determination of penalty parameters in quantum annealing. Quantum Sci.

Technol. 11, 015039, DOI: 10.1088/2058-9565/ae30a2 (2026).

11. Rusnáková, R., Chovanec, M. & Gazda, J. Quantum annealing for minimum bisection problem: A machine learning-based

approach for penalty parameter tuning (2025). 2509.19005.

12. Roch, C., Ratke, D., Nüßlein, J., Gabor, T. & Feld, S. The effect of penalty factors of constrained hamiltonians on the

eigenspectrum in quantum annealing. ACM Transactions on Quantum Comput. 4, 1–18, DOI: 10.1145/3577202 (2023).

13. Jeong, S.-G., Cong, M. D., Noh, D.-I., Pham, Q.-V. & Hwang, W.-J. Embedding-aware noise modeling of quantum

annealing (2025). 2510.04594.

Factorization machine with quadratic-optimization annealing (FMQA)[1,2] is a representative method for black-box optimization (BBO), which aims to find a set of decision variables that minimizes (or maximizes) an objective function using only observed input–output data. In BBO, surrogate model-based methods are widely employed to obtain high-quality solutions under a limited number of expensive function evaluations. FMQA combines a machine learning model, the factorization machine (FM), with an Ising machine.

Because FMQA searches for candidate solutions using an Ising machine, integer variables must be discretized into binary variables. In this study, we use one-hot encoding for this discretization, in which each variable is represented by multiple bits, with exactly one bit taking the value of 1 and the others taking the value of 0. However, the training dataset for the FM in FMQA often exhibits an imbalance in the occurrence of bits taking the value of 1, and some bits may take the value of 0 for all data points in the dataset. As a result, the corresponding FM parameters are not updated, which can reduce the estimation accuracy of the FM and, consequently, degrade the optimization performance of FMQA. Therefore, it is important to ensure sufficient bit diversity by making each bit take the value of 1 at least once in the dataset.

In this study, we focus on the initial training dataset and aim to improve optimization performance by addressing insufficient bit diversity. To this end, we propose an extended FMQA method that generates the initial training data using quasi-random sequences, such as Latin hypercube sampling and Sobol’ sequences, so that each bit takes the value 1 at least once in the dataset.

In this presentation, we evaluate the effectiveness of the proposed method on the human-powered aircraft benchmark problem [3], which involves the design of the main wing. The results show that, under the same number of objective function evaluations, the proposed method achieves better objective values than conventional FMQA, which uses uniform random sampling for initial data generation. Furthermore, the proposed method achieves better objective values than Bayesian optimization and NSGA-II, which are representative classical optimization algorithms.
[1] K. Kitai et al., Phys. Rev. Res. 2, 013319 (2020).

[2] R. Tamura et al., Applied Physics Reviews 13, 021307 (2026).

[3] N. Namura, in Evol. Multi-Criterion Optim., pp. 195–210 (Springer, Singapore, 2025).

Designing evacuation routes is essential for minimizing evacuation time and preventing congestion. This is a combinatorial optimization problem where the computational cost increases significantly with the size of the building. In this study, we use Quantum Annealing (QA) to demonstrate its utility in selecting appropriate routes more quickly than exact method, based on the distribution of people within the building.

We modeled evacuation in a multi-story building by dividing people into groups. We developed a QUBO model to minimize the time required for the last group to finish evacuation. The model includes two main constraints: each group must choose exactly one route, and each path segment can only be used by one group at a time to avoid congestion.

Our results demonstrate that QA effectively distributes groups across different routes and minimizes the final time. While classical exact solvers provided better solution quality, QA was significantly faster. Given the trade-off between solution quality and calculation time, QA shows potential as a tool for providing real-time evacuation guidance during emergencies.
Masayuki Ohzeki. "Reconsideration of optimization for reduction of traffic congestion." Journal of the Physical Society of Japan 93.115003 (2024)

Coherent analog quantum computing with superconducting qubits places stringent demands on processor characterization and control, requiring careful mitigation of parasitic couplings, signal distortions, and decoherence mechanisms before meaningful quantum experiments can be performed. This talk presents advances at Qilimanjaro on a fluxonium-based processor implemented in a flip-chip architecture, building on the platform introduced at AQC 2025. The architecture and its key design choices for coherent analog operation are reviewed, providing context for the experimental results that follow. These include crosstalk characterization across a multi-qubit subsystem [1] and flux-pulse distortion compensation [2], which together establish the control foundation for coherent operation. Finally, single-qubit analog schedules are discussed, outlining how the fluxonium energy landscape can be programmatically traversed to implement coherent analog protocols. Taken together, these results represent a significant step in validating fluxonium flip-chip processors as a viable platform for coherent analog quantum information processing.

[1] Dai, X et al., Calibration of Flux Crosstalk in Large-Scale Flux-Tunable Superconducting Quantum Circuits, PRX Quantum 2, 040313, 20 October, 2021

[2] Hellings, Christoph et al “Calibrating magnetic flux control in superconducting circuits by compensating distortions on timescales from nanoseconds up to tens of microseconds.” Physical Review Research 7, 043142 (2025)

Quantum annealing currently represents one of the most mature quantum computing technologies, making it a promising platform for exploring near-term applications in probabilistic machine learning. In this work, we investigate the use of a quantum annealer as a proposal generator within a Markov Chain Monte Carlo framework for Bayesian inference in machine learning models.

A classical neural network is first pretrained and subsequently frozen, while predictive uncertainty is introduced through a Bayesian treatment of the final-layer weights. The corresponding posterior over weights is expressed as a QUBO problem, allowing candidate weight configurations to be generated on the D-Wave quantum annealer. These hardware-generated samples are used to construct a Markov Chain utilizing the Metropolis-Hastings algorithm. In the process acceptance probabilities correct for proposal bias and drive convergence toward the target posterior distribution.

To study scaling behavior, the Bayesian treatment is extended from the final layer to multi-layer architectures, increasing posterior complexity and introducing non-convex structures in which classical local proposal mechanisms may become inefficient. In this regime, quantum annealing is investigated as a source of non-local proposals capable of improving posterior exploration.

A central challenge is that conditional proposal probabilities are not directly accessible. We therefore discuss practical strategies for real-device implementation: empirical estimation of transition probabilities through repeated anneals and approximate symmetry assumptions for reverse-annealing transitions.

Finally, we analyze how quantum-generated proposals affect posterior exploration and uncertainty estimation, and discuss their potential for uncertainty-aware machine learning applications in which reliable posterior inference is critical.

This study proposes a quantum annealing–based optimization approach for determining optimal crop planting layouts in heterogeneous mountainous farmland.

Due to limited flat land in South Korea, cultivation often occurs in mountainous areas with highly variable environmental conditions.

To address this, each plot is assigned a suitability score, and planting decisions are formulated as quadratic unconstrained binary optimization problem that incorporates spacing constraints and crop quantity requirements. This optimization problem is then solved utilizing quantum annealing.

Additionally, planting intervals are modeled using both rectangular and elliptical approaches, and their performance is compared. Based on this analysis, we identify a more effective method for real-world agricultural environments and propose an optimal planting strategy that balances crop yield and quality under constrained farmland conditions.

his study addresses the transportation routing problem for elderly services in a university-affiliated hospital located in eastern Taiwan by developing an optimization model that simultaneously considers vehicle capacity constraints, caregiver allocation, time window requirements, and directional service constraints. The problem setting involves 42 elderly participants, 2 vehicles, and 7 caregivers, and is formulated as a vehicle routing problem with capacity and time windows, which is further transformed into a Quadratic Unconstrained Binary Optimization (QUBO) model and solved using a quantum annealing approach. The objective is to minimize the total travel distance and service violations while satisfying operational constraints, including mandatory arrival at the day-care center between 08:30 and 09:30 and return trips scheduled between 16:00 and 17:00. By integrating practical service considerations with advanced optimization techniques, the proposed framework demonstrates the effectiveness of quantum annealing in solving complex, real-world routing problems and provides valuable insights for transportation scheduling and smart mobility planning in long-term care systems, particularly in geographically dispersed regions such as eastern Taiwan.

Variational autoencoders (VAEs) learn compact latent representations of complex data, but their generative capacity is fundamentally constrained by the choice of prior distribution over the latent space. Energy-based priors offer a principled way to move beyond factorized assumptions and capture structured interactions among latent variables, yet training such priors at scale requires accurate and efficient sampling from intractable distributions. Here we present Boltzmann-machine--prior VAEs (BM-VAEs) trained using quantum annealing--based sampling in three distinct operational modes within a single generative system. During training, diabatic quantum annealing (DQA) provides unbiased Boltzmann samples for gradient estimation of the energy-based prior; for unconditional generation, slower quantum annealing (QA) concentrates samples near low-energy minima; for conditional generation, bias fields are added to direct sampling toward attribute-specific regions of the energy landscape (c-QA). Using up to 2000 qubits on a D-Wave Advantage2 processor, we demonstrate stable and efficient training across multiple datasets, with faster convergence and lower reconstruction loss than a Gaussian-prior VAE. The learned Boltzmann prior enables unconditional generation by sampling directly from the energy-based latent distribution, a capability that plain autoencoders lack, and conditional generation through latent biasing that leverages the learned pairwise interactions.

https://arxiv.org/abs/2604.00919

Chia-Ho Ou1,2, Cheng-Yu Wu2, and Feng-Yu Lin2

1Graduate School of Information Sciences, Tohoku University, Sendai, Japan

2Dept. of Computer Science and Information Engineering, National Pingtung University, Taiwan

cho@mail.nptu.edu.tw

As Taiwan transitions toward a nuclear-free, low-carbon power system [1], determining the optimal genera- tion mix across multiple energy sources while balancing cost, emissions, and supply stability is a large-scale combinatorial optimization problem. Conventional approaches such as mixed-integer linear programming often fail to scale when the problem involves discrete generation levels, competing objectives, and numer- ous policy constraints.

We reformulate this problem as a Quadratic Unconstrained Binary Optimization (QUBO) [2] model that encodes multiple energy sources and combines economic, environmental, and stability objectives into a single formulation. We further introduce a bounded-slack mechanism that relaxes the strict supply–demand equality while retaining its physical interpretation.

The QUBO model represents eight energy types (renewables, coal, oil, natural gas, cogeneration, pumped hydro, nuclear, and battery storage) using one-hot encoding over discrete generation levels, with nuclear fixed at zero to reflect Taiwan’s post-decommissioning policy [3]. The objective function combines cost, emissions, and instability as a weighted sum, where the higher stability weight reflects national policy emphasis on grid reliability [4]. Policy constraints, including renewable share requirements, fossil fuel caps, and a renewable–battery coupling term [5], are incorporated as soft quadratic penalties following the Ising–QUBO equivalence [6]. For the supply–demand balance, a bounded-slack formulation provides a physical reserve-margin interpretation, paired with a soft deviation penalty to guide solvers toward demand- matching solutions.

We first evaluate the formulation at a large scale experiment, comparing four solvers: Compal GPU Anneal- ing (CGA), Greedy local search, Simulated Annealing (SA), and Simulated Quantum Annealing (SQA). Ex- periments are conducted across multiple demand levels anchored at Taiwan’s historical generation baseline of 2,524 billion kWh [7]. All solvers consistently produce feasible solutions without constraint violations, which validates the bounded-slack formulation. Among them, CGA achieves the best objective values across all demand levels.

At a reduced small-scale, we extend the comparison to six solvers by adding QAOA via Qiskit [8] with the Aer simulator, and CUDA-Q QAOA [9] with GPU-accelerated simulation. Both QAOA variants produce feasible solutions across all demand levels; however, their objective values remain approximately 5–15% higher than those of CGA and classical heuristics, a result consistent with shallow QAOA circuit depth on structured QUBO problems [10].

This work provides a cross-paradigm QUBO benchmark for energy dispatch, where classical, quantum- inspired, and quantum approaches are compared on a policy-grounded problem instance. Results show that the bounded-slack formulation improves feasibility, while a clear performance gap remains between current quantum algorithms and classical methods at this problem scale.

References

[1] Executive Yuan, Taiwan, “Energy portfolio and electricity pricing policy report,” Oct. 2024.

[2] F. Glover, G. Kochenberger, and Y. Du, “A tutorial on formulating and using QUBO models,” arXiv preprint

arXiv:1811.11538, 2019.

[3] Taiwan Power Company, “Maanshan Unit 2 decommissioning and diversified dispatch planning,” 2025.

[4] Bureau of Energy, MOEA, Energy Statistics Handbook 2024, 2024.

[5] National Development Council, Taiwan, “Taiwan’s pathway to net-zero emissions in 2050,” 2022.

[6] A. Lucas, “Ising formulations of many NP problems,” Frontiers in Physics, vol. 2, p. 5, 2014.

[7] Taiwan Power Company, “Historical generation mix statistics,” 2025.

[8] E. Farhi, J. Goldstone, and S. Gutmann, “A quantum approximate optimization algorithm,” arXiv preprint

arXiv:1411.4028, 2014.

[9] NVIDIA, “CUDA-Q: A platform for hybrid quantum-classical computing,” 2024.

[10] L. Zhou et al., “Quantum approximate optimization algorithm: Performance, mechanism, and implementation

on near-term devices,” Physical Review X, vol. 10, no. 2, p. 021067, 2020.

     Solving the NP-hard trajectory optimization problem for 6-DOF industrial robots presents a profound computational challenge for both classical numerical methods and near-term quantum devices. While classical sampling-based algorithms (e.g., RRT) are widely used, they face severe computational bottlenecks and poor path smoothness in narrow passages. Conversely, pure quantum annealing approaches are currently restricted by qubit scalability when processing high-dimensional kinematics.

     In this study, we propose a pragmatic hybrid quantum-classical framework designed to bridge this hardware limitation. By delegating the computationally demanding Inverse Kinematics (IK) and global rough routing to classical solvers, our algorithm effectively isolates localized bottleneck segments. The core contribution of this work lies in the quantum formulation: we formulate these critical segments into Constrained Quadratic Models (CQMs). Unlike traditional unconstrained models (QUBO) that require complex penalty-weight tuning, our CQM approach distinctly separates objective functions (minimizing path length and joint variations) from absolute hard constraints (strict motor limits and collision-free boundaries). Implemented using the D-Wave Ocean SDK, we utilize simulated annealing to solve this quantum-inspired formulation, strictly enforcing these physical constraints to achieve optimal path smoothing in dense obstacle regions. We benchmark our hybrid model against pure classical RRT across varying 3D complexities, demonstrating a highly scalable, safe, and efficient framework for real-world industrial trajectory optimization. 

Trajectory Optimization: Cutting-Edge Trajectory Optimization through Quantum Annealing (MDPI). This paper demonstrates the feasibility of transforming continuous trajectory planning into quantum-solvable models, serving as the foundational motivation for our advanced CQM approach.

Inverse Kinematics: Quantum Annealing for Inverse Kinematics in Robotics. This work explores the translation of complex kinematic trigonometric functions into quantum formats, which supports our rationale for isolating IK and handling it classically within a hybrid framework.

Path Planning & Collision Avoidance: Scalable Multi-Robot Path Planning via Quadratic Unconstrained Binary Optimization. This recent publication highlights the trend of using quantum optimization (QUBO) for spatial planning; our work builds upon this by employing CQMs to natively handle collision-free hard constraints without complex penalty tuning.

With the growing demand for effective self-directed learning, the scheduling of personal learning tasks has become increasingly important. This paper proposes a solution based on the quantum-inspired approach using the Compal GPU Annealer (CGA) to optimize personal learning schedules and compare its effectiveness with traditional greedy and genetic algorithms. A mathematical model is proposed in the paper, considering several task priorities, deadlines, and daily hour limits. Besides, concepts from Cognitive Load Theory are applied to handle human-centric constraints, balancing task difficulty to prevent psychological overload. The proposed problem is converted into a Quadratic Unconstrained Binary Optimization (QUBO) form for quantum-inspired computing. Experimental results indicate that the proposed approach effectively maximizes task completion efficiency while maintaining cognitive stability.

As competition in deep-sea fishing intensifies, route planning has shifted from experience-driven decisions to computational optimization under strict operational constraints. In practice, fuel consumption is not static: it is shaped by spatial ocean currents, time-varying sea states, and increasing payload as catches accumulate. These coupled dynamics create a planning problem that goes beyond classical profit-based capacitated routing, requiring models that are both physically grounded and computationally scalable. This work proposes a Quadratic Unconstrained Binary Optimization (QUBO) formulation for dynamic fishery routing using time-indexed binary variables to represent vessel location over discrete operational steps. The objective integrates catch revenue with a dynamic fuel-cost term in which baseline travel cost is multiplicatively modulated by (i) a current-field factor derived from simulated vortices and (ii) sea-state multipliers that vary by time step, while a load-amplification factor captures the additional fuel required as payload approaches the vessel’s capacity. Operational rules including port-to-port departure/return, one-location-per-step, and at-most-once zone visits are enforced via quadratic penalties. We evaluate the approach on simulated instances and solve the resulting QUBOs using the Compal GPU Annealer (CGA), a quantum-inspired annealing platform leveraging GPU-parallel energy-based search. The results demonstrate that the proposed model reliably produces feasible, capacity-compliant routes while capturing fuel variations induced by changing marine conditions and payload accumulation.

We formulate the elementary school teacher scheduling problem in Taiwan as a Quadratic Unconstrained Binary Optimization (QUBO) model and solve it using Digital Annealing (DA). Unlike Western scheduling benchmarks, Taiwanese elementary schools feature unique constraints such as homeroom-teacher-based curricula in grades 1–2, mother-tongue language instruction by specialist teachers, fixed school-wide professional development time slots, and itinerant subject teachers. We evaluate DA against Simulated Annealing (SA) on real-world instances from a 30-class elementary school, comparing feasibility rate, solution quality, and time-to-target across varying problem scales. This work contributes the first QUBO formulation of Taiwanese elementary school scheduling and provides empirical insights into digital annealing's applicability in educational scheduling.

Parametric Hamiltonians can exhibit point-like spectrum degeneracies (diabolic points or conical intersections), which can cause singularities in the quantum metric tensor of eigenstate manifolds. We show how these singularities can be regularised by a coordinate transformation and introduce a formalism, in which diabolic points act as "bridges" between adjacent eigenstate manifolds. When multiple diabolic points are present, we introduce the connected state manifold, characterise its topology, and refine the rules for nodal lines. We demonstrate the applicability of this framework in numerical stability near degeneracies, geodesic shortcuts via diabolic points, and Berry phase calculations for paths traversing diabolic points.

To understand the thermal limitations of open-system quantum annealers, which are expected to relax toward a thermal Boltzmann distribution, evaluating the effective machine temperature β is crucial. However, conventional thermometry methods typically neglect the freeze-out phenomenon during the annealing process, introducing estimation biases into β. In this work, we propose an annealing framework incorporating the freeze-out mechanism and analytically derive the exact solutions for the expectation and variance of the nearest-neighbor correlation function under the one-dimensional Ising model. Since our proposed model is governed by two parameters —the freeze-out time s* and the effective temperature β— we attempt to simultaneously estimate both parameters using the method of moments based on the exact expectation and variance. Our analysis reveals that while sufficient estimation accuracy is achieved in certain parameter regions, other regions suffer from estimation failures due to a structural degeneracy. Nevertheless, we verify that the estimated parameters successfully correspond to the “effective” freeze-out point and temperature that optimally account for the output distribution within our model framework, suggesting that our simultaneous estimation approach serves as a more robust evaluation tool compared to conventional methods that disregard the freeze-out phenomenon.