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

Graph Neural Network Autoencoders for Efficient Quantum Circuit Optimisation

Henri Ahola

Published May 22, 2026• 0 citations
OpenAlex

Abstract

Successfully harnessing the power of present-day quantum computing hardware requires efficient utilisation of the limited available resources. Most notably, in the quantum circuit model for computation, circuit depth is severely limited by noisy equipment. Techniques to optimise and reduce depth are consequently necessary. Various approaches exist, but many suffer from poor scalability in practice. In a search for better techniques, this thesis extends preliminary research efforts to implement a quantum circuit compressor or optimiser that utilises a graph variational autoencoder to manipulate circuits in a computationally more manageable latent space. Such latent spaces aim to capture and compress relevant information into far fewer dimensions. This reduces the computational cost of search space exploration and sidesteps the combinatorial explosion that traditional optimisers typically face. As an intermediate implementation and demonstration of the technique, the work presented in this thesis does not yet optimise circuits; it does, however, successfully compress chosen subsets of quantum circuits into a low-dimensional space with a near-100% perfect reconstruction rate. Further work would extend this to additionally manipulate—and eventually optimise—the compressed circuits.

Related papers

Graph Neural Network Autoencoders for Efficient Quantum Circuit Optimisation

Ioana Moflic, Vikas Garg, Alexandru Paler

Reinforcement learning (RL) is a promising method for quantum circuit optimisation. However, the state space that has to be explored by an RL agent is extremely large when considering all the possibilities in which a quantum circuit can be transformed through local rewrite operations. This state space explosion slows down the learning of RL-based optimisation strategies. We present for the first time how to use graph neural network (GNN) autoencoders for the optimisation of quantum circuits. We construct directed acyclic graphs from the quantum circuits, encode the graphs and use the encodings to represent RL states. We illustrate our proof of concept implementation on Bernstein-Vazirani circuits and, from preliminary results, we conclude that our autoencoder approach: a) maintains the optimality of the original RL method; b) reduces by 20 \% the size of the table that encodes the learned optimisation strategy. Our method is the first realistic first step towards very large scale RL quantum circuit optimisation.

arXiv:2303.03280•Mar 6, 2023•5 citations

AltGraph: Redesigning Quantum Circuits Using Generative Graph Models for Efficient Optimization

Collin Beaudoin, Koustubh Phalak, Swaroop Ghosh

Quantum circuit transformation aims to optimize circuits for depth, gate count, and compatibility with Noisy Intermediate Scale Quantum (NISQ) devices that suffer from various error sources. Prior methods use combinations of expert-defined rules and Reinforcement Learning (RL). We introduce AltGraph, a novel approach employing generative graph models to generate functionally equivalent quantum circuits using—specifically, Direct Acyclic Graph (DAG) Variational Autoencoder (D-VAE) variants (GRU and GCN) and Deep Generative Model for Graphs (DeepGMG). AltGraph perturbs the latent space to generate quantum circuits optimized for hardware coupling maps, reducing gate count by 37.55% and circuit depth by 37.75% post-transpiling, with 0.0074 Mean Squared Error (MSE) in the density matrix—outperforming state-of-the-art methods by 2.56%.

arXiv:2403.12979•ACM Great Lakes Symposium on VLSI (GLSVLSI)•Jun 10, 2024•9 citations

Generative graph neural networks for inverse design of cQED circuits

Valerio Brugnami

Superconducting quantum circuits are among the most promising platforms for quantum computing, but their design remains a complex inverse problem, requiring the identification of circuit architectures and physical parameters capable of reproducing a desired quantum response. Existing approaches mainly rely on iterative optimization, predefined topologies, or previously simulated device families, limiting their flexibility as circuit complexity increases. This thesis presents a generative framework for the inverse design of superconducting quantum circuits. The proposed approach combines a Graph Variational Autoencoder with Transformer-based generative architectures, including GPT-like autoregressive and BERT-like bidirectional mechanisms, to learn a shared latent representation between circuit graphs and their effective Hamiltonians. Starting from a target Hamiltonian, the framework generates both the circuit topology and its continuous physical parameters, while physics-based constraints and latent-space optimization are used to improve the quality of the generated solutions. The results show accurate reconstruction of known circuit families, a high fraction of physically valid generated circuits, and encouraging generalization to previously unseen topologies. These results indicate the potential of modern generative AI techniques for the automated inverse design of superconducting quantum circuits. i

•Oct 29, 2026•0 citations

Guided graph compression for quantum graph neural networks: application to jet tagging at CERN's LHC

Casals Baena, Mikel

Graph Neural Networks (GNNs) are effective for analyzing graph-structured data but face challenges with large graphs due to high memory requirements and inefficient sparse matrix operations on GPUs. Quantum Computing (QC) may be used to overcome these problems. For instance, Quantum Graph Neural Networks (QGNNs) have already been implemented in the literature. However, current quantum hardware limits the dimensionality of the datasets that can be encoded. For this reason, previous works simplify graph datasets manually or use small synthetic datasets. This work introduces the Guided Graph Compression (GGC) framework. It uses a graph autoencoder to reduce the size of graphs in terms of number of nodes and node feature dimensionality. The compression is guided towards enhancing the performance of a classification task by coupling the loss functions of the autoencoder and the classifier. The framework is applied to the Jet Tagging task that is relevant at CERN's LHC. It involves classifying jets as being originated from quarks or gluons. It is tested on three different training paradigms. First, a classical GNN is trained using the uncompressed data. Then, graph autoencoders are trained as a separate preprocessing step and used for classical and quantum classifiers. Lastly, the GGC framework is applied to both types of classifiers. Results show that the GGC framework outperforms the other two paradigms. Moreover, the classical and quantum classifiers achieve competitive performance in terms of ROC-AUC score.

•Sep 3, 2024•0 citations

Neural Architecture Search for Quantum Autoencoders

Hibah Agha, Samuel Yen-Chi Chen, Huan-Hsin Tseng +1 more

In recent years, machine learning and deep learning have driven advances in domains such as image classification, speech recognition, and anomaly detection by leveraging multi-layer neural networks to model complex data. Simultaneously, quantum computing (QC) promises to address classically intractable problems via quantum parallelism, motivating research in quantum machine learning (QML). Among QML techniques, quantum autoencoders show promise for compressing high-dimensional quantum and classical data. However, designing effective quantum circuit architectures for quantum autoencoders remains challenging due to the complexity of selecting gates, arranging circuit layers, and tuning parameters. This paper proposes a neural architecture search (NAS) framework that automates the design of quantum autoencoders using a genetic algorithm (GA). By systematically evolving variational quantum circuit (VQC) configurations, our method seeks to identify high-performing hybrid quantum-classical autoencoders for data reconstruction without becoming trapped in local minima. We demonstrate effectiveness on image datasets, highlighting the potential of quantum autoencoders for efficient feature extraction within a noise-prone, near-term quantum era. Our approach lays a foundation for broader application of genetic algorithms to quantum architecture search, aiming for a robust, automated method that can adapt to varied data and hardware constraints.

quant-pharXiv:2511.19246•Nov 25, 2025•0 citations

Learning Efficient Circuits for Hamiltonian Simulation

Joseph Richard Gibbs

Quantum computers promise an exciting path to the scalable simulation of quantum matter. However, in their nascent state, the accessible gate depth of quantum computation on available hardware remains highly constrained. In this thesis, I present a suite of hybrid quantum-classical algorithms that aid Hamiltonian simulation across both the Noisy Intermediate-Scale Quantum era and future fault-tolerant architectures. By treating classical computation as an active resource to refine quantum algorithms, I develop automated, variational methods to learn hardware-efficient circuits. This work aims to overcome traditional trainability barriers, such as barren plateaus, by embedding physical symmetries directly into the circuit design. Central to the thesis is the synergistic use of tensor networks to classically pre-compute weakly entangled stages of a quantum simulation. This enables the compression of deep circuits, drastically reducing both the accumulation of gate errors on near-term physical qubits and the prohibitive overhead of operations like magic state distillation on error-corrected hardware, reserving the quantum processor strictly for the most challenging, highly entangled dynamics. These methodologies are ultimately directed toward the ab-initio simulation of subatomic physics, providing a scalable approach for the first-principles study of strongly correlated systems, including nuclear structure and lattice gauge theories.

•Jul 31, 2026•0 citations
Successfully harnessing the power of present-day quantum computing hardware requires efficient utilisation of the limited available resources. Most notably, in the quantum circuit model for computation, circuit depth is severely limited by noisy equipment. Techniques to optimise and reduce depth are consequently necessary. Various approaches exist, but many suffer from poor scalability in practice. In a search for better techniques, this thesis extends preliminary research efforts to implement a quantum circuit compressor or optimiser that utilises a graph variational autoencoder to manipulate circuits in a computationally more manageable latent space. Such latent spaces aim to capture and compress relevant information into far fewer dimensions. This reduces the computational cost of search space exploration and sidesteps the combinatorial explosion that traditional optimisers typically face. As an intermediate implementation and demonstration of the technique, the work presented in this thesis does not yet optimise circuits; it does, however, successfully compress chosen subsets of quantum circuits into a low-dimensional space with a near-100% perfect reconstruction rate. Further work would extend this to additionally manipulate—and eventually optimise—the compressed circuits.
Reinforcement learning (RL) is a promising method for quantum circuit optimisation. However, the state space that has to be explored by an RL agent is extremely large when considering all the possibilities in which a quantum circuit can be transformed through local rewrite operations. This state space explosion slows down the learning of RL-based optimisation strategies. We present for the first time how to use graph neural network (GNN) autoencoders for the optimisation of quantum circuits. We construct directed acyclic graphs from the quantum circuits, encode the graphs and use the encodings to represent RL states. We illustrate our proof of concept implementation on Bernstein-Vazirani circuits and, from preliminary results, we conclude that our autoencoder approach: a) maintains the optimality of the original RL method; b) reduces by 20 \% the size of the table that encodes the learned optimisation strategy. Our method is the first realistic first step towards very large scale RL quantum circuit optimisation.
Quantum circuit transformation aims to optimize circuits for depth, gate count, and compatibility with Noisy Intermediate Scale Quantum (NISQ) devices that suffer from various error sources. Prior methods use combinations of expert-defined rules and Reinforcement Learning (RL). We introduce AltGraph, a novel approach employing generative graph models to generate functionally equivalent quantum circuits using—specifically, Direct Acyclic Graph (DAG) Variational Autoencoder (D-VAE) variants (GRU and GCN) and Deep Generative Model for Graphs (DeepGMG). AltGraph perturbs the latent space to generate quantum circuits optimized for hardware coupling maps, reducing gate count by 37.55% and circuit depth by 37.75% post-transpiling, with 0.0074 Mean Squared Error (MSE) in the density matrix—outperforming state-of-the-art methods by 2.56%.
Superconducting quantum circuits are among the most promising platforms for quantum computing, but their design remains a complex inverse problem, requiring the identification of circuit architectures and physical parameters capable of reproducing a desired quantum response. Existing approaches mainly rely on iterative optimization, predefined topologies, or previously simulated device families, limiting their flexibility as circuit complexity increases. This thesis presents a generative framework for the inverse design of superconducting quantum circuits. The proposed approach combines a Graph Variational Autoencoder with Transformer-based generative architectures, including GPT-like autoregressive and BERT-like bidirectional mechanisms, to learn a shared latent representation between circuit graphs and their effective Hamiltonians. Starting from a target Hamiltonian, the framework generates both the circuit topology and its continuous physical parameters, while physics-based constraints and latent-space optimization are used to improve the quality of the generated solutions. The results show accurate reconstruction of known circuit families, a high fraction of physically valid generated circuits, and encouraging generalization to previously unseen topologies. These results indicate the potential of modern generative AI techniques for the automated inverse design of superconducting quantum circuits. i
Graph Neural Networks (GNNs) are effective for analyzing graph-structured data but face challenges with large graphs due to high memory requirements and inefficient sparse matrix operations on GPUs. Quantum Computing (QC) may be used to overcome these problems. For instance, Quantum Graph Neural Networks (QGNNs) have already been implemented in the literature. However, current quantum hardware limits the dimensionality of the datasets that can be encoded. For this reason, previous works simplify graph datasets manually or use small synthetic datasets. This work introduces the Guided Graph Compression (GGC) framework. It uses a graph autoencoder to reduce the size of graphs in terms of number of nodes and node feature dimensionality. The compression is guided towards enhancing the performance of a classification task by coupling the loss functions of the autoencoder and the classifier. The framework is applied to the Jet Tagging task that is relevant at CERN's LHC. It involves classifying jets as being originated from quarks or gluons. It is tested on three different training paradigms. First, a classical GNN is trained using the uncompressed data. Then, graph autoencoders are trained as a separate preprocessing step and used for classical and quantum classifiers. Lastly, the GGC framework is applied to both types of classifiers. Results show that the GGC framework outperforms the other two paradigms. Moreover, the classical and quantum classifiers achieve competitive performance in terms of ROC-AUC score.
In recent years, machine learning and deep learning have driven advances in domains such as image classification, speech recognition, and anomaly detection by leveraging multi-layer neural networks to model complex data. Simultaneously, quantum computing (QC) promises to address classically intractable problems via quantum parallelism, motivating research in quantum machine learning (QML). Among QML techniques, quantum autoencoders show promise for compressing high-dimensional quantum and classical data. However, designing effective quantum circuit architectures for quantum autoencoders remains challenging due to the complexity of selecting gates, arranging circuit layers, and tuning parameters. This paper proposes a neural architecture search (NAS) framework that automates the design of quantum autoencoders using a genetic algorithm (GA). By systematically evolving variational quantum circuit (VQC) configurations, our method seeks to identify high-performing hybrid quantum-classical autoencoders for data reconstruction without becoming trapped in local minima. We demonstrate effectiveness on image datasets, highlighting the potential of quantum autoencoders for efficient feature extraction within a noise-prone, near-term quantum era. Our approach lays a foundation for broader application of genetic algorithms to quantum architecture search, aiming for a robust, automated method that can adapt to varied data and hardware constraints.
Quantum computers promise an exciting path to the scalable simulation of quantum matter. However, in their nascent state, the accessible gate depth of quantum computation on available hardware remains highly constrained. In this thesis, I present a suite of hybrid quantum-classical algorithms that aid Hamiltonian simulation across both the Noisy Intermediate-Scale Quantum era and future fault-tolerant architectures. By treating classical computation as an active resource to refine quantum algorithms, I develop automated, variational methods to learn hardware-efficient circuits. This work aims to overcome traditional trainability barriers, such as barren plateaus, by embedding physical symmetries directly into the circuit design. Central to the thesis is the synergistic use of tensor networks to classically pre-compute weakly entangled stages of a quantum simulation. This enables the compression of deep circuits, drastically reducing both the accumulation of gate errors on near-term physical qubits and the prohibitive overhead of operations like magic state distillation on error-corrected hardware, reserving the quantum processor strictly for the most challenging, highly entangled dynamics. These methodologies are ultimately directed toward the ab-initio simulation of subatomic physics, providing a scalable approach for the first-principles study of strongly correlated systems, including nuclear structure and lattice gauge theories.