Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Log inSign up
Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Blog
Job Board
Q3AS

© 2026 Aqora Quantum S.A.S.

TermsPrivacyLegal Notice

arXiv:2609.19392

Transformers as Intrinsic Optimizers for Quantum Approximate Optimization Algorithm

Kuan-Cheng Chen, Xiaotian Xu, Hiromichi Matsuyama, Wei-Hao Huang, Haomu Yuan, Yu Yamashiro

quant-phSubmitted Sep 16, 2026• Announced Sep 18, 2026• 0 citations
arXivHTMLPDF

Abstract

The Quantum Approximate Optimization Algorithm (QAOA) is a leading variational framework for combinatorial optimization on noisy intermediate-scale quantum hardware, but its practical performance depends strongly on the classical optimizer used to train its variational parameters. This outer-loop optimization is often nonconvex, initialization-sensitive, and costly when repeated across large families of related problem instances. In this work, we propose a Transformer-based intrinsic optimization framework for QAOA, in which the optimizer itself is learned and embedded directly into the hybrid quantum-classical loop. The proposed graph-conditioned Transformer processes problem structure, current QAOA parameters, measurement feedback, and recent optimization history to predict the next variational-parameter update, thereby reformulating instance-wise classical optimization as an amortized learned policy. We develop a mathematical formulation of this intrinsic-optimization perspective and evaluate the method on QAOA-based MaxCut benchmarks across multiple problem settings, with comparisons against representative classical and learned optimization baselines. The results demonstrate that Transformer-based intrinsic optimization can provide a structured and transferable mechanism for improving the classical component of hybrid quantum optimization algorithms.

AI Summary

The paper asks whether a learned, graph-conditioned optimizer can amortize QAOA parameter training across related MaxCut instances. It uses a Transformer to predict angle updates from a graph embedding, current angles, previous updates, objective feedback, and optimization progress. One token represents each QAOA layer, with an additional graph-context token. Updates are bounded by a scaled $\tanh$ and clipped to the parameter domain. Meta-training optimizes performance along unrolled QAOA trajectories, with optional update regularization and imitation pretraining. Evaluation uses disjoint training, validation, and test graph sets, comparing against MLP and LSTM learned optimizers, Adam variants, COBYLA, Nelder–Mead, and SPSA. At depth $p=2$, the Transformer’s mean expected-cut approximation ratios are $0.8932$, $0.8766$, and $0.8767$ for $n=5,8,11$ vertices, respectively. Corresponding MLP ratios are $0.8923$, $0.8709$, and $0.8708$; multi-start Adam achieves $0.8914$, $0.8716$, and $0.8662$. Additional studies vary graph size up to 13 qubits and depth up to $p=10$. For six-qubit, depth-four instances, reported transfer performance approaches saturation with approximately 50 training graphs. A model-size sweep reports its highest ratios around $0.94$ with moderate-capacity Transformers. The stated scope is small MaxCut instances and primarily same-distribution transfer. With vectorized-adjacency inputs, results apply within fixed-size regimes; variable-size use requires a compatible graph encoder. Larger and heterogeneous graph families, out-of-distribution transfer, hardware execution, hardware-aware noise and cost features, and theoretical analysis of learned dynamics are left for future work.

Likely largely AI-generatedIncrementalSmall numerics onlyUnclear noveltyOverclaims

Central claim may not hold: The claimed matched-budget comparison and optimization-efficiency benefits cannot be verified because actual baseline evaluation budgets, gradient costs, stopping criteria, and offline training costs are not reported. The Transformer exceeds the MLP's reported mean approximation ratios by only 0.0009, 0.0057, and 0.0059, and the benchmark does not provide enough information to quantify uncertainty in these differences. The supplied source contains no plotted data for the figure-based experiments, leaving their claimed transfer, scaling, depth, model-size, and ablation trends supported only by narrative descriptions. The zero-update ablation does not establish that graph conditioning or measurement/history feedback contributes to performance, because it compares learned updates only against leaving the initial parameters unchanged. The initialization-noise experiment is not reproducibly specified or explicitly reconciled with the methodology's zero-angle initialization.

Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.

Related papers

Quantum Approximate Optimization Algorithm (QAOA) for Solving NP-Hard Problems

Srikanth Kilaru, Thota Sravya Vardhani, Attluri Udaya Lakshmi +2 more

The Quantum Approximate Optimization Algorithm (QAOA) is a hybrid quantum-classical algorithm used to study NP-hard combinatorial optimization problems. In this work, we focused on optimization strategies for QAOA including circuit depth reduction, parameter parallelization, and hybrid classical feedback mechanisms. We will consider the application of QAOA to different types of NP-hard problems such as Max-Cut and Vehicle Routing Problem (VRP) and discuss the advantages and disadvantages of each in terms of approximation quality and cost of computation. The comparative simulation on the various qubit counts and depth layers points out the limits of scalability and the constraints of hardware on the current Noisy Intermediate-Scale Quantum (NISQ) devices. And also an improved divide-and-conquer strategy is presented in which parameter reusability and gradient-informed initialization are combined to boost convergence rates. The results of the experiment show that it is more efficient and scalable while maintaining the competitive approximation performance compared to the baseline QAOA. The experimental validation focuses on representative small-size instances, while the proposed framework provides extensions for scalable implementations. Ultimately, this work is a step towards the realization of efficient, hardware-aware, and scalable optimization strategies, thus opening up the practical deployment of QAOA on the quantum systems of the next few years.

IEEE Nexo-Tech International Conference on Advanced Technologies and Innovations (ICATI)•May 8, 2026•0 citations

Learning to Optimize Variational Quantum Circuits to Solve Combinatorial Problems

Sami Khairy, Ruslan Shaydulin, Łukasz Cincio +2 more

Quantum computing is a computational paradigm with the potential to outperform classical methods for a variety of problems. Proposed recently, the Quantum Approximate Optimization Algorithm (QAOA) is considered as one of the leading candidates for demonstrating quantum advantage in the near term. QAOA is a variational hybrid quantum-classical algorithm for approximately solving combinatorial optimization problems. The quality of the solution obtained by QAOA for a given problem instance depends on the performance of the classical optimizer used to optimize the variational parameters. In this paper, we formulate the problem of finding optimal QAOA parameters as a learning task in which the knowledge gained from solving training instances can be leveraged to find high-quality solutions for unseen test instances. To this end, we develop two machine-learning-based approaches. Our first approach adopts a reinforcement learning (RL) framework to learn a policy network to optimize QAOA circuits. Our second approach adopts a kernel density estimation (KDE) technique to learn a generative model of optimal QAOA parameters. In both approaches, the training procedure is performed on small-sized problem instances that can be simulated on a classical computer; yet the learned RL policy and the generative model can be used to efficiently solve larger problems. Extensive simulations using the IBM Qiskit Aer quantum circuit simulator demonstrate that our proposed RL- and KDE-based approaches reduce the optimality gap by factors up to 30.15 when compared with other commonly used off-the-shelf optimizers.

arXiv:1911.11071•Proceedings of the AAAI Conference on Artificial Intelligence•Apr 3, 2020•129 citations

Extending QAOA-GPT to Higher-Order Quantum Optimization Problems

Leanto Sunny, Abhinav Rijal, George Siopsis

The recently proposed QAOA-GPT framework demonstrated that generative pre-trained transformers can learn mappings between problem graphs and optimized quantum circuits for the Quantum Approximate Optimization Algorithm (QAOA). In this work, we extend QAOA-GPT to Higher-Order Unconstrained Binary Optimization (HUBO) problems, focusing on spin-glass Hamiltonians that include cubic interaction terms. Using FEATHER graph embeddings to encode topological information, we train the model on graph-circuit pairs generated via ADAPT-QAOA and evaluate its performance on 8- and 16-qubit instances embedded on heavy-hex lattices. The generative model produces adaptive QAOA-like circuits and corresponding variational parameters in a single forward pass, bypassing the iterative classical optimization loop. The generated circuits achieve average approximation ratios exceeding 0.95, closely matching classically optimized ADAPT-QAOA results, while maintaining consistent parameter distributions across circuit depths. These results demonstrate that QAOA-GPT generalizes effectively to higher-order cost Hamiltonians and complex energy landscapes, establishing generative modeling as a scalable pathway toward autonomous variational circuit design and quantum algorithm discovery in the NISQ era.

quant-pharXiv:2511.07391•Nov 11, 2025•4 citations

CVAE-Guided QAOA Parameter Initialization and Dynamic Injection

Jai Nana, Xinpeng Li, Ning Xie +7 more

The quantum approximate optimization algorithm (QAOA) is a hybrid quantum-classical variational algorithm designed to tackle combinatorial optimization problems that are classically intractable. Careful parameter initialization in QAOA can reduce the number of quantum circuit evaluations required, making it more feasible to run on noisy intermediate-scale quantum (NISQ) devices. In this work, We focus on MaX-Cut, a classic NP-hard optimization problem, and propose a conditional variational autoencoder (CVAE) scheme for generating initial QAOA parameters for the Max-Cut problems based on input graph structures. The CVAE, trained on optimal parameters from 270 ten-node graphs, produces multiple parameter candidates conditioned on 27-dimensional feature vectors that incorporate spectral properties as well as learned embeddings. We evaluate and compare two strategies: direct initialization using the highestconfidence prediction, and dynamic injection which monitors optimization progress and intervenes when needed. On 89 test graphs, the dynamic injection strategy increased successful convergence cases from 41 to 49, reduced mean iteration count by 4.94.

IEEE International Conference on Quantum Computing and Engineering (QCE)•Aug 30, 2025•0 citations

Accelerating Quantum Approximate Optimization Algorithm using Machine Learning

Mahabubul Alam, Abdullah Ash Saki, Swaroop Ghosh

We propose a machine learning based approach to accelerate quantum approximate optimization algorithm (QAOA) implementation which is a promising quantum-classical hybrid algorithm to prove the so-called quantum supremacy. In QAOA, a parametric quantum circuit and a classical optimizer iterates in a closed loop to solve hard combinatorial optimization problems. The performance of QAOA improves with increasing number of stages (depth) in the quantum circuit. However, two new parameters are introduced with each added stage for the classical optimizer increasing the number of optimization loop iterations. We note a correlation among parameters of the lower-depth and the higher-depth QAOA implementations and, exploit it by developing a machine learning model to predict the gate parameters close to the optimal values. As a result, the optimization loop converges in a fewer number of iterations. We choose graph MaxCut problem as a prototype to solve using QAOA. We perform a feature extraction routine using 100 different QAOA instances and develop a training data-set with 13, 860 optimal parameters. We present our analysis for 4 flavors of regression models and 4 flavors of classical optimizers. Finally, we show that the proposed approach can curtail the number of optimization iterations by on average 44.9% (up to 65.7%) from an analysis performed with 264 flavors of graphs.

arXiv:2002.01089•Design, Automation, and Test in Europe•Mar 1, 2020•56 citations

Quantum-Assisted Hybrid Optimization for Graph-Based Combinatorial Optimization

Srinivasarao Thota, N Shilpa, Nasr Al Din Ide

Graph-based combinatorial optimization problems are computationally challenging for classical optimization techniques due to their NP-hard nature. This paper proposes a novel Quantum-Assisted Hybrid Optimization Algorithm (QAHOA) that integrates the Quantum Approximate Optimization Algorithm (QAOA), spectral graph theory, and classical optimization strategies for graph-based combinatorial optimization. The developed framework uses spectral initialization via the graph Laplacian Fiedler vector to provide structurally informed starting variational parameters, aiming to stabilize the parameter search space and improve convergence trajectories under noisy conditions. The mathematical framework, alongside a convergence analysis to first-order stationary points, is provided to establish theoretical baseline properties. Evaluated on synthetic Erdős–Rényi random graph instances, the current study serves as a preliminary, small-to-medium-scale proof-of-concept. Controlled evaluations demonstrate that the proposed QAHOA achieves marginal yet stable and statistically significant improvements in mean cut values and approximation ratios over classical Greedy heuristics, Simulated Annealing, the standard QAOA, and Warm-Start QAOA. Formal paired hypothesis testing mathematically verifies the stability of the framework, while preliminary 4-qubit hardware executions demonstrate baseline architectural compatibility under physical NISQ noise channels.

IEEE Access•Jan 1, 2026•0 citations
The Quantum Approximate Optimization Algorithm (QAOA) is a leading variational framework for combinatorial optimization on noisy intermediate-scale quantum hardware, but its practical performance depends strongly on the classical optimizer used to train its variational parameters. This outer-loop optimization is often nonconvex, initialization-sensitive, and costly when repeated across large families of related problem instances. In this work, we propose a Transformer-based intrinsic optimization framework for QAOA, in which the optimizer itself is learned and embedded directly into the hybrid quantum-classical loop. The proposed graph-conditioned Transformer processes problem structure, current QAOA parameters, measurement feedback, and recent optimization history to predict the next variational-parameter update, thereby reformulating instance-wise classical optimization as an amortized learned policy. We develop a mathematical formulation of this intrinsic-optimization perspective and evaluate the method on QAOA-based MaxCut benchmarks across multiple problem settings, with comparisons against representative classical and learned optimization baselines. The results demonstrate that Transformer-based intrinsic optimization can provide a structured and transferable mechanism for improving the classical component of hybrid quantum optimization algorithms.
The paper asks whether a learned, graph-conditioned optimizer can amortize QAOA parameter training across related MaxCut instances. It uses a Transformer to predict angle updates from a graph embedding, current angles, previous updates, objective feedback, and optimization progress. One token represents each QAOA layer, with an additional graph-context token. Updates are bounded by a scaled tanh⁡\tanhtanh and clipped to the parameter domain. Meta-training optimizes performance along unrolled QAOA trajectories, with optional update regularization and imitation pretraining.
Evaluation uses disjoint training, validation, and test graph sets, comparing against MLP and LSTM learned optimizers, Adam variants, COBYLA, Nelder–Mead, and SPSA. At depth p=2p=2p=2, the Transformer’s mean expected-cut approximation ratios are 0.89320.89320.8932, 0.87660.87660.8766, and 0.87670.87670.8767 for n=5,8,11n=5,8,11n=5,8,11 vertices, respectively. Corresponding MLP ratios are 0.89230.89230.8923, 0.87090.87090.8709, and 0.87080.87080.8708; multi-start Adam achieves 0.89140.89140.8914, 0.87160.87160.8716, and 0.86620.86620.8662. Additional studies vary graph size up to 13 qubits and depth up to p=10p=10p=10. For six-qubit, depth-four instances, reported transfer performance approaches saturation with approximately 50 training graphs. A model-size sweep reports its highest ratios around 0.940.940.94 with moderate-capacity Transformers.
The stated scope is small MaxCut instances and primarily same-distribution transfer. With vectorized-adjacency inputs, results apply within fixed-size regimes; variable-size use requires a compatible graph encoder. Larger and heterogeneous graph families, out-of-distribution transfer, hardware execution, hardware-aware noise and cost features, and theoretical analysis of learned dynamics are left for future work.
The Quantum Approximate Optimization Algorithm (QAOA) is a hybrid quantum-classical algorithm used to study NP-hard combinatorial optimization problems. In this work, we focused on optimization strategies for QAOA including circuit depth reduction, parameter parallelization, and hybrid classical feedback mechanisms. We will consider the application of QAOA to different types of NP-hard problems such as Max-Cut and Vehicle Routing Problem (VRP) and discuss the advantages and disadvantages of each in terms of approximation quality and cost of computation. The comparative simulation on the various qubit counts and depth layers points out the limits of scalability and the constraints of hardware on the current Noisy Intermediate-Scale Quantum (NISQ) devices. And also an improved divide-and-conquer strategy is presented in which parameter reusability and gradient-informed initialization are combined to boost convergence rates. The results of the experiment show that it is more efficient and scalable while maintaining the competitive approximation performance compared to the baseline QAOA. The experimental validation focuses on representative small-size instances, while the proposed framework provides extensions for scalable implementations. Ultimately, this work is a step towards the realization of efficient, hardware-aware, and scalable optimization strategies, thus opening up the practical deployment of QAOA on the quantum systems of the next few years.
Quantum computing is a computational paradigm with the potential to outperform classical methods for a variety of problems. Proposed recently, the Quantum Approximate Optimization Algorithm (QAOA) is considered as one of the leading candidates for demonstrating quantum advantage in the near term. QAOA is a variational hybrid quantum-classical algorithm for approximately solving combinatorial optimization problems. The quality of the solution obtained by QAOA for a given problem instance depends on the performance of the classical optimizer used to optimize the variational parameters. In this paper, we formulate the problem of finding optimal QAOA parameters as a learning task in which the knowledge gained from solving training instances can be leveraged to find high-quality solutions for unseen test instances. To this end, we develop two machine-learning-based approaches. Our first approach adopts a reinforcement learning (RL) framework to learn a policy network to optimize QAOA circuits. Our second approach adopts a kernel density estimation (KDE) technique to learn a generative model of optimal QAOA parameters. In both approaches, the training procedure is performed on small-sized problem instances that can be simulated on a classical computer; yet the learned RL policy and the generative model can be used to efficiently solve larger problems. Extensive simulations using the IBM Qiskit Aer quantum circuit simulator demonstrate that our proposed RL- and KDE-based approaches reduce the optimality gap by factors up to 30.15 when compared with other commonly used off-the-shelf optimizers.
The recently proposed QAOA-GPT framework demonstrated that generative pre-trained transformers can learn mappings between problem graphs and optimized quantum circuits for the Quantum Approximate Optimization Algorithm (QAOA). In this work, we extend QAOA-GPT to Higher-Order Unconstrained Binary Optimization (HUBO) problems, focusing on spin-glass Hamiltonians that include cubic interaction terms. Using FEATHER graph embeddings to encode topological information, we train the model on graph-circuit pairs generated via ADAPT-QAOA and evaluate its performance on 8- and 16-qubit instances embedded on heavy-hex lattices. The generative model produces adaptive QAOA-like circuits and corresponding variational parameters in a single forward pass, bypassing the iterative classical optimization loop. The generated circuits achieve average approximation ratios exceeding 0.95, closely matching classically optimized ADAPT-QAOA results, while maintaining consistent parameter distributions across circuit depths. These results demonstrate that QAOA-GPT generalizes effectively to higher-order cost Hamiltonians and complex energy landscapes, establishing generative modeling as a scalable pathway toward autonomous variational circuit design and quantum algorithm discovery in the NISQ era.
The quantum approximate optimization algorithm (QAOA) is a hybrid quantum-classical variational algorithm designed to tackle combinatorial optimization problems that are classically intractable. Careful parameter initialization in QAOA can reduce the number of quantum circuit evaluations required, making it more feasible to run on noisy intermediate-scale quantum (NISQ) devices. In this work, We focus on MaX-Cut, a classic NP-hard optimization problem, and propose a conditional variational autoencoder (CVAE) scheme for generating initial QAOA parameters for the Max-Cut problems based on input graph structures. The CVAE, trained on optimal parameters from 270 ten-node graphs, produces multiple parameter candidates conditioned on 27-dimensional feature vectors that incorporate spectral properties as well as learned embeddings. We evaluate and compare two strategies: direct initialization using the highestconfidence prediction, and dynamic injection which monitors optimization progress and intervenes when needed. On 89 test graphs, the dynamic injection strategy increased successful convergence cases from 41 to 49, reduced mean iteration count by 4.94.
We propose a machine learning based approach to accelerate quantum approximate optimization algorithm (QAOA) implementation which is a promising quantum-classical hybrid algorithm to prove the so-called quantum supremacy. In QAOA, a parametric quantum circuit and a classical optimizer iterates in a closed loop to solve hard combinatorial optimization problems. The performance of QAOA improves with increasing number of stages (depth) in the quantum circuit. However, two new parameters are introduced with each added stage for the classical optimizer increasing the number of optimization loop iterations. We note a correlation among parameters of the lower-depth and the higher-depth QAOA implementations and, exploit it by developing a machine learning model to predict the gate parameters close to the optimal values. As a result, the optimization loop converges in a fewer number of iterations. We choose graph MaxCut problem as a prototype to solve using QAOA. We perform a feature extraction routine using 100 different QAOA instances and develop a training data-set with 13, 860 optimal parameters. We present our analysis for 4 flavors of regression models and 4 flavors of classical optimizers. Finally, we show that the proposed approach can curtail the number of optimization iterations by on average 44.9% (up to 65.7%) from an analysis performed with 264 flavors of graphs.
Graph-based combinatorial optimization problems are computationally challenging for classical optimization techniques due to their NP-hard nature. This paper proposes a novel Quantum-Assisted Hybrid Optimization Algorithm (QAHOA) that integrates the Quantum Approximate Optimization Algorithm (QAOA), spectral graph theory, and classical optimization strategies for graph-based combinatorial optimization. The developed framework uses spectral initialization via the graph Laplacian Fiedler vector to provide structurally informed starting variational parameters, aiming to stabilize the parameter search space and improve convergence trajectories under noisy conditions. The mathematical framework, alongside a convergence analysis to first-order stationary points, is provided to establish theoretical baseline properties. Evaluated on synthetic Erdős–Rényi random graph instances, the current study serves as a preliminary, small-to-medium-scale proof-of-concept. Controlled evaluations demonstrate that the proposed QAHOA achieves marginal yet stable and statistically significant improvements in mean cut values and approximation ratios over classical Greedy heuristics, Simulated Annealing, the standard QAOA, and Warm-Start QAOA. Formal paired hypothesis testing mathematically verifies the stability of the framework, while preliminary 4-qubit hardware executions demonstrate baseline architectural compatibility under physical NISQ noise channels.