Skip to content

Aqasch/awesome-QAS

BranchesTags

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

51 Commits
assets
assets
 
 
CITATION.cff
CITATION.cff
 
 
README.md
README.md
 
 

Repository files navigation

Awesome QAS (Quantum Architecture Search)

Awesome Open Source Love

A curated list of standout libraries, tutorials, research papers, and essential resources focused on Quantum Architecture Search (QAS). This collection is designed to serve as a structured and thorough reference, empowering researchers and developers to accelerate their work and stay at the forefront of QAS advancements. 🔼 Last updated: 4/3/26

QAS

Table of Contents

Quantum Architecture Search

Quantum Architecture Search (QAS) is an automated framework that combines machine learning and optimization techniques to design noise-resilient quantum circuits with minimal depth and gate counts, tailored to specific computational tasks and hardware constraints. This process addresses the critical challenge of balancing expressivity (the ability to model complex problems) with practical limitations like quantum noise and hardware connectivity.

Papers

  • Automated Design of Quantum Circuits (1998), one of the earliest paper in quantum architecture search, proposes an automated method for designing quantum circuits using genetic programming, a type of evolutionary algorithm inspired by natural selection. Unlike traditional approaches that rely on manual design or exhaustive enumeration of all possible gate sequences, the authors introduce a flexible, stochastic search framework capable of discovering efficient quantum circuits with minimal prior knowledge of the target unitary operation. The algorithm evolves circuit candidates by applying operations such as mutation, crossover, and transposition, and evaluates their fitness based on how closely the resulting unitary matches a desired transformation. Demonstrated on the quantum teleportation task, the method consistently finds valid and sometimes novel circuit designs more efficiently than brute-force search.

  • Quantum optimization with a novel Gibbs objective function and ansatz architecture search (2020) modifies the Quantum Approximate Optimization Algorithm (QAOA) through two innovations: (1) a Gibbs objective function, prioritizing low-energy state discovery over energy expectation, and (2) ansatz architecture search (AAS), automating circuit design to optimize gate counts and performance.

  • DQAS: Differentiable quantum architecture search (2021) A general framework automates quantum circuit design through end-to-end differentiable optimization, combining differentiable programming (gradient-based tuning), probabilistic programming (architectural uncertainty handling), and quantum programming (practical implementation). It demonstrates versatility by decomposing unitary operations into gates, mitigating noise in circuits, and optimizing layouts for combinatorial problems (e.g., QAOA). This framework accelerates the discovery of quantum advantages in NISQ-era devices while bridging interdisciplinary insights, offering a scalable tool for both practical applications and theoretical exploration in quantum computing.

  • Quantum architecture search via deep reinforcement learning (2021) automates quantum circuit design using deep reinforcement learning (DRL), addressing the challenge of creating efficient gate sequences for target states (e.g., multi-qubit GHZ states) without expert knowledge. The DRL agent, trained via A2C and PPO algorithms, accesses only Pauli-X/Y/Z expectation values and predefined quantum operations to optimize gate synthesis. This framework eliminates the need for embedded quantum physics principles in the agent, demonstrating generality for diverse DRL architectures and gate compilation studies, enabling resource-efficient quantum circuit development.

  • Neural predictor based quantum architecture search (2021) automates the design of parameterized quantum circuits (PQCs) for variational algorithms, inspired by classical AutoML. A neural predictor evaluates circuit performance with 10× fewer evaluations than random search, achieving state-of-the-art results in quantum simulation and machine learning. Optimal PQCs and predictors are transferable across tasks, combining quantum computing with classical efficiency tools to accelerate practical quantum advantage.

  • Quantum Architecture Search with Meta-Learning (2021) accelerates quantum circuit design by leveraging meta-learning to transfer prior task experiences, avoiding inefficient scratch-based searches. It pre-trains meta-architectures and gate parameters across tasks, enabling rapid adaptation to new problems (e.g., variational compiling, QAOA) with fewer gradient updates. Simulations show MetaQAS converges faster and achieves better solutions than gradient-based QAS (e.g., DQAS) after fine-tuning, demonstrating its potential to streamline quantum algorithm deployment in NISQ-era applications.

  • Quantum Architecture Search via Continual Reinforcement Learning (2021) enhances quantum circuit design by integrating continual learning to address changing device noise, unlike prior DRL methods requiring frequent retraining. By reusing learned policies across noise environments, PPR-DQL accelerates the discovery of quantum gate sequences (e.g., generating two-qubit Bell states faster than training from scratch) while reducing resource overhead.

  • Quantum circuit optimization with deep reinforcement learning (2021) the key objective is to address the gap in current optimization methods that overlook hardware-specific details, an essential consideration for near-term quantum devices. The authors employ a deep convolutional neural network as an agent to autonomously learn and apply generic strategies to optimize arbitrary circuits on a specific architecture.

  • Reinforcement learning for optimization of variational quantum circuit architectures (2021) an autonomous algorithm that balances circuit expressivity and depth constraints for near-term quantum devices. Using feedback-driven curriculum learning, the method dynamically adjusts task complexity based on real-time performance, incrementally refining ground-state energy estimates while minimizing circuit depth.

  • Quantum circuit architecture search for variational quantum algorithms (2022) enhances variational quantum algorithms (VQAs) by automatically optimizing ansatz design to balance expressivity and noise resilience in NISQ-era devices. While larger ansatze boost expressivity, they risk noise accumulation and poor trainability (e.g., barren plateaus). QAS efficiently identifies near-optimal circuit architectures, tested via IBM quantum hardware and simulators for data classification and quantum chemistry tasks. Results show QAS outperforms manual ansatz selection, mitigating noise, improving trainability, and achieving higher accuracy—demonstrating its potential to unlock robust quantum advantages in practical applications.

  • Evolutionary quantum architecture search for parametrized quantum circuits (2022) automates the design of parameterized quantum circuits (PQCs) for hybrid quantum-classical reinforcement learning (RL) systems using a population-based genetic algorithm. By evolving architectures through mutations (e.g., adding/deleting gates) and crossover operations, it explores the space of quantum operations (e.g., variational layers, entanglement, data encoding) to optimize performance on RL benchmarks like CartPole and MountainCar.

  • Search space pruning for quantum architecture search (2022) addresses the exponential search space challenge in automating quantum circuit design for variational algorithms (VQAs) by progressively eliminating low-potential gates. Using rotation-based indicators (superior to cost-based metrics with limited parameter updates), the method filters unpromising candidates early, reducing computational overhead.

  • Quantum Circuit Architecture Search on a Superconducting Processor (2022) to enhance Variational Quantum Algorithms (VQAs) on an 8-qubit superconducting quantum processor. The researchers tailor the hardware-efficient ansatz towards classification tasks using QAS, which leads to a significant improvement in test accuracy compared to heuristic ansatze.

  • Evolutionary-based quantum architecture search (2022)tThis scheme aims to balance higher expressive power and trainability in error-prone and depth-limited quantum circuits in the Noisy Intermediate-Scale Quantum (NISQ) era. EQAS encodes quantum circuit layouts into binary strings (quantum genes) and employs an algorithm to remove redundant parameters using the eigenvalues of the quantum Fisher information matrix.

  • QuantumNAS: Noise-Adaptive Search for Robust Quantum Circuits (2022) aims to make quantum circuits more resilient to noise in Noisy Intermediate-Scale Quantum (NISQ) computers. The authors introduce a novel SuperCircuit to decouple the circuit search and parameter training, which improves the efficiency of finding the best variational circuit and its optimal parameters.

  • A GNN-based predictor for quantum architecture search (2023) accelerates circuit design for variational quantum algorithms (VQAs) by using a graph neural network (GNN) predictor to estimate circuit performance without full evaluations. Representing circuits as directed acyclic graphs (DAGs), a GNN encoder with asynchronous message-passing maps structures to performance features, mimicking quantum computations. Tested on variational quantum eigensolver (VQE) tasks, the predictor identifies high-performing circuits, filters weak candidates, and boosts sample efficiency—dramatically reducing computational costs in QAS.

  • QuantumDARTS: Differentiable Quantum Architecture Search for Variational Quantum Algorithms (2023) employs Gumbel-Softmax optimization to automate parameterized quantum circuit (PQC) design for variational algorithms (e.g., Max-Cut, energy estimation, image classification), bypassing costly circuit sampling. The framework features macro search, which constructs full circuits with minimal prior knowledge, and micro search, an innovative method that infers scalable sub-circuit patterns from small-scale problems for large-scale tasks.

  • EQNAS: Evolutionary Quantum Neural Architecture Search for Image Classification (2023) automates quantum circuit design for quantum neural networks (QNNs), addressing manual design challenges like high complexity and low classification accuracy. Using a quantum evolutionary algorithm, EQNAS initializes a population of circuits encoded with quantum image representations, evaluates their fitness (e.g., classification performance), and iteratively updates them via quantum rotation gates, circuit construction, and entirety interference crossover.

  • GSQAS: Graph Self-supervised Quantum Architecture Search (2023) enhances quantum architecture search (QAS) by addressing the scarcity of labeled training circuits through self-supervised learning.

  • Quantum architecture search via truly proximal policy optimization (2023) enhances deep reinforcement learning-based QAS by addressing limitations in policy ratio constraints and trust region enforcement from prior Proximal Policy Optimization (PPO) methods. It introduces a rollback clipping function to strictly bound the probability ratio between old and new policies, preventing destabilizing deviations during training. Simultaneously, trust region constraints (via KL-divergence) ensure policies remain within optimization-safe zones, guaranteeing monotonic performance improvements.

  • Quantum Reinforcement Learning for Quantum Architecture Search (2023) automates multi-qubit GHZ state preparation using the asynchronous advantage actor-critic (A3C) algorithm, building on prior DRL-based QAS methods. The agent, devoid of pre-encoded quantum knowledge, interacts with a predefined gate set and observes only Pauli-X/Y/Z expectation values to iteratively construct circuits.

  • Differentiable Quantum Architecture Search for Quantum Reinforcement Learning (2023) demonstrates potential in advancing quantum reinforcement learning (QRL) by automating circuit design for quantum deep Q-learning tasks. The study investigates DQAS’s ability to adaptively construct task-specific quantum circuits across diverse datasets, leveraging differentiable optimization to dynamically adjust circuit architectures based on learning objectives.

  • Training-Free Quantum Architecture Search (2024) improves circuit design by replacing resource-intensive training with two proxies: a path-based metric (counting paths in circuit DAGs) filters weak candidates at zero cost, followed by an expressibility-based proxy to pinpoint high-performance circuits.

  • Distributed quantum architecture search (2024) automates circuit design across interconnected quantum processing units (QPUs), addressing NISQ-era qubit limits. The framework integrates TeleGate and TeleData methods to implement nonlocal gates via entanglement and classical communication, while optimizing qubit assignment from logical to physical layouts. A two-stage training-free strategy first filters circuits using a zero-cost path-based proxy (counting DAG paths) and then refines with an expressibility metric, reducing evaluation costs.

  • Curriculum reinforcement learning for quantum architecture search under hardware errors (2024) addresses noise-aware circuit design by integrating a 3D tensor-based encoding (capturing gate types, qubits, and circuit depth) with reinforcement learning to efficiently explore the architecture space. The method employs an episode halting scheme to prioritize shorter circuits and a modified simultaneous perturbation stochastic approximation (SPSA) optimizer for noise-resilient parameter tuning, accelerating convergence. A Pauli-transfer matrix (PTM) simulator in the Pauli-Liouville basis enables efficient noisy circuit simulations, combining gate operations with hardware error models.

  • Differentiable Quantum Architecture Search in Asynchronous Quantum Reinforcement Learning (2024) streamlines quantum reinforcement learning (QRL) by automating the design of variational quantum circuits (VQCs) through gradient-based optimization, which jointly trains circuit parameters and architectural weights. It tackles challenges in data encoding and parameterized circuit design, critical for QRL performance, while asynchronous reinforcement learning enables parallel training to boost efficiency.

  • KANQAS:Kolmogorov-Arnold Network for Quantum Architecture Search (2024) addresses limitations of traditional MLP-based QAS methods by leveraging the Kolmogorov-Arnold theorem to design quantum circuits with improved interpretability and efficiency. KANs replace MLPs’ fixed activation functions with spline-based learnable univariate functions, reducing the number of parameters while enhancing adaptability.

  • Quantum Machine Learning Architecture Search via Deep Reinforcement Learning (2024) a novel approach to designing effective quantum machine learning (QML) models using deep reinforcement learning (RL). The authors address the challenge of balancing model complexity and feasibility on Noisy Intermediate-Scale Quantum (NISQ) devices by employing RL to discover proficient QML model architectures tailored for specific supervised learning tasks.

  • Application of ZX-calculus to quantum architecture search (2024) by combining ZX-calculus techniques with Genetic Programming (GP) to optimize parameterized quantum circuits used in Quantum Machine Learning (QML). The primary objective is to address the challenges in designing efficient quantum circuits for QML tasks.

  • A quantum information theoretic analysis of reinforcement learning-assisted quantum architecture search (2024) the authors investigate RL-QAS, specifically for variational quantum state diagonalisation problem. The study analyzes various dimensions such as entanglement thresholds, the impact of initial conditions on RL-agent performance, phase transition behavior of correlations, and the discrete contributions of qubits in deducing eigenvalues through conditional entropy.

  • Reinforcement learning-based architecture search for quantum machine learning (2024) address the challenge of heuristic architecture selection by employing a model-based reinforcement learning algorithm, which reduces the number of necessary circuit evaluations and provides a sample-efficient framework. The researchers utilize a layered circuit structure to significantly reduce the search space and account for multiple objectives, such as solution quality, hardware restrictions, and circuit depth.

  • SA-DQAS: Self-attention Enhanced Differentiable Quantum Architecture Search (2024) with a self-attention mechanism to optimize the design of quantum circuits for Quantum Machine Learning (QML) challenges. This approach treats quantum circuits as sequences of placeholders containing quantum gates, leveraging self-attention to capture dependencies among operations, unlike the independent placement in DQAS.

  • An RNN-policy gradient approach for quantum architecture search (2024) uses deep reinforcement learning to automatically design quantum circuit architectures. The main objective is to find the optimal quantum circuit composition architecture for a given task, enhancing the performance capability of quantum algorithms. The approach involves learning the sampling of the circuit architecture through a reinforcement learning-based controller and applying layer-based search to accelerate computational efficiency.

  • Trainability maximization using estimation of distribution algorithms assisted by surrogate modelling for quantum architecture search (2024) employing an online surrogate model and a novel metric called information content to address the challenges of high measurement costs and barren plateaus in quantum architecture search.

  • Quantum Circuit Optimization with AlphaTensor (2024) is a deep reinforcement learning-based method designed to optimize quantum circuits by minimizing the number of costly T gates required for fault-tolerant quantum computation. By leveraging the connection between T-count optimization and tensor decomposition, this approach can incorporate domain-specific quantum knowledge and utilize specialized gadgets, leading to significant reductions in T-count compared to traditional methods.

  • Quantum circuit synthesis with diffusion models (2024) leverages generative machine learning, specifically denoising diffusion models (DMs), to bridge the gap between abstract quantum operations and their physical implementation in quantum circuits. By employing text-conditioning, the method directs the DM to generate quantum operations tailored for gate-based circuits, effectively bypassing the exponential computational cost of simulating quantum dynamics during training.

  • Challenges for Reinforcement Learning in Quantum Circuit Design (2024) propose a generic RL framework, formalized as a Markov decision process, that enables agents to learn low-level quantum control by selecting and parameterizing universal quantum gates to solve two main tasks: state preparation and unitary composition. They introduce the Quantum Circuit Designer (QCD), a customizable RL environment built to benchmark various RL algorithms, including PPO, A2C, TD3, and SAC, across a range of quantum objectives. Through comprehensive evaluations, the study highlights the limitations of current RL methods in exploring sparse reward landscapes and high-dimensional continuous action spaces, while showing SAC's superior performance due to its effective exploration strategies.

  • Quantum Architecture Search with Unsupervised Representation Learning (2025) It decouples circuit representation learning (via an improved graph encoding scheme) from the search process, eliminating reliance on labeled data. Unsupervised training captures structural patterns in quantum circuits, generating a latent space where REINFORCE or Bayesian Optimization efficiently explores high-performing architectures.

  • Benchmarking Quantum Architecture Search with Surrogate Assistance (2025) addresses the challenges of automating parameterized quantum circuit (PQC) design for variational algorithms (VQAs) by introducing a standardized benchmarking framework. Quantum architecture search (QAS) aims to optimize PQC architectures for performance and hardware compatibility, but progress is hindered by the lack of uniform evaluation methods and high computational costs. SQuASH leverages surrogate models—trained on pre-collected quantum circuit performance data—to accelerate and standardize QAS evaluations.

  • Differentiable Quantum Architecture Search in Quantum-Enhanced Neural Network Parameter Generation (2025) to address the challenges of designing effective quantum circuit architectures in the Quantum-Train (QT) framework for quantum-enhanced neural network parameter generation.

  • Topology-Driven Quantum Architecture Search Framework (2025) to address the challenges of computational complexity in existing Quantum Architecture Search (QAS) algorithms. The framework first identifies optimal circuit topologies using QAS and then fine-tunes gate types with an efficient QAS that inherits parameters from the topology search phase.

  • Self-supervised representation learning for Bayesian quantum architecture search (2025) addresses the computational bottlenecks of traditional QAS by integrating a graph isomorphism network (GIN) predictor pretrained on circuit expressibility—a measure of a quantum circuit’s ability to uniformly explore Hilbert space—to create a structured latent representation of architectures . This self-supervised pretraining reduces reliance on labeled data while improving predictor generalizability

  • Scalable Quantum Architecture Search via Landscape Analysis (2025) that efficiently explores and evaluates quantum circuits via landscape fluctuation analysis. The framework's main objective is to address the trade-off between trainability and expressibility in variational quantum computing by predicting circuit learnability without costly training. The authors achieve this by combining landscape fluctuation analysis with a streamlined two-level search strategy, which enables the identification of high-performance, large-scale circuits with fewer gates and higher accuracy.

  • TensorRL-QAS: Reinforcement learning with tensor networks for scalable quantum architecture search (2025) the main objective is to address the scalability issues faced by RL-based quantum architecture search (QAS) methods, which encounter significant computational and training costs as the number of qubits, circuit depth, and noise increase. By warm-starting the architecture search with a matrix product state approximation of the target solution, TensorRL-QAS effectively narrows the search space and accelerates convergence to the desired solution. The framework is tested on several quantum optimization problems of up to 20-qubit.

  • DeQompile: quantum circuit decompilation using genetic programming for explainable quantum architecture search (2025) a novel genetic programming-based decompiler framework to address the challenge of explainability in quantum architecture search (QAS). The framework, implemented in the open-source tool DeQompile, employs program synthesis techniques, including symbolic regression and abstract syntax tree manipulation, to reverse-engineer high-level quantum algorithms from low-level circuit representations.

  • RhoDARTS: Differentiable Quantum Architecture Search with Density Matrix Simulations (2025) models the search process as the evolution of a quantum mixed state, emerging from the search space of quantum architectures. The algorithm is designed to address the challenge of identifying effective Quantum Neural Network (QNN) architectures for general machine learning tasks by approaching QAS from a quantum perspective.

  • Reinforcement learning with learned gadgets to tackle hard quantum problems on real hardware (2025) integrates reinforcement learning with program synthesis to design quantum circuits for complex tasks. By incorporating composite gates (gadgets) into the action space, GRL enhances the exploration of parameterized quantum circuits (PQCs) for tasks such as approximating ground states of quantum Hamiltonians.

  • Scaling the Automated Discovery of Quantum Circuits via Reinforcement Learning with Gadgets (2025) presents a novel approach to enhancing the scalability of reinforcement learning in quantum circuit design. The main objective is to address the challenges associated with RL's limited scalability due to the exponential increase in computation times with growing circuit complexity. The authors utilize the concept of "gadgets" which are composite gates that can be incorporated into RL frameworks to improve efficiency and enable the discovery of highly complex quantum codes.

  • Improving thermal state preparation of Sachdev-Ye-Kitaev model with reinforcement learning on quantum hardware (2025) addresses the significant challenge of preparing thermal states for large systems of the Sachdev-Ye-Kitaev (SYK) model on near-term quantum processors, which is complicated by the rapid increase in complexity of parameterized quantum circuits as system size grows. The authors propose a scalable framework that integrates reinforcement learning with convolutional neural networks, iteratively optimizing quantum circuits and their parameters using a composite reward signal based on entropy and the expectation values of the SYK Hamiltonian.

  • QAS-QTNs: Curriculum Reinforcement Learning-Driven Quantum Architecture Search for Quantum Tensor Networks (2025) follows the developments of CRLQAS, applying curriculum reinforcement learning-driven quantum architecture search and additionally extends performance assessment across multiple RL agents and strategies. This includes benchmarking both classical and quantum-enhanced agents for robust circuit optimization as complexity increases.

  • QuProFS: An Evolutionary Training-free Approach to Efficient Quantum Feature Map Search (2025) is a training-free, evolutionary quantum architecture search that ranks circuit feature maps via fast proxy metrics (like expressivity and robustness), achieving competitive accuracy and up to two times faster search than existing QAS approaches without costly parameter training, on both quantum simulators and hardware.

  • Quantum Long Short-term Memory with Differentiable Architecture Search (2025) is an end-to-end framework that jointly optimizes quantum circuit architecture and parameters for sequence learning tasks. It consistently outperforms handcrafted QLSTM baselines by adapting circuit design during training, achieving lower loss and better generalization on time-series and NLP benchmarks.

  • Quantum Architecture Search for Solving Quantum Machine Learning Tasks (2025) introduces RL for quantum architecture search to discover effective circuit architectures for classification tasks. We evaluate RL-QAS using the Iris and binary MNIST datasets. The agent autonomously discovers low-complexity circuit designs that achieve high test accuracy. Our results show that RL is a viable approach for automated architecture search in quantum machine learning.

  • Neural Architecture Search Algorithms for Quantum Autoencoders (2025) introduces two quantum neural architecture search (NAS) algorithms that automatically design parameter-efficient quantum autoencoder circuits for data compression tasks. The authors model circuits as repeatable patterns and combine evolutionary search with reinforcement learning to discover encoder architectures that generalize across tasks.

  • Hybrid action Reinforcement Learning for quantum architecture search (2025) introduces HyRLQAS, a hybrid-action reinforcement learning framework that simultaneously learns quantum gate placement and parameter initialization for variational quantum circuits used in VQE. By treating structure (discrete gate choices) and parameters (continuous values) within a single policy, the agent reuses optimization experience and provides informed warm-starts instead of relying solely on external optimizers that restart from scratch.

  • QASER: Breaking the Depth vs. Accuracy Trade-Off for Quantum Architecture Search (2025) introduces QASER, a reinforcement learning–based quantum architecture search method that uses a carefully engineered exponential reward function to jointly optimize circuit depth and accuracy. By explicitly encoding both energy error and resource costs such as depth and two-qubit gate count into the reward, QASER steers the RL agent toward circuits that are simultaneously shallow and precise for quantum chemistry state preparation.

  • Neural Architecture Search for Quantum Autoencoders (2025) presents a neural architecture search framework that uses a genetic algorithm to automatically design variational quantum circuits for quantum autoencoders. By evolving gate types, layer configurations, and circuit parameters, the method searches over hybrid quantum–classical autoencoder architectures to maximize reconstruction quality while avoiding local minima.

  • Distributed quantum architecture search using multi-agent reinforcement learning (2025) introduces a multi-agent reinforcement learning framework for quantum architecture search, where each agent controls a local block of a parameterized quantum circuit rather than using a single global agent. This distributed design reduces the dimensionality of each agent’s action space, leading to faster convergence and lower computational cost when searching for problem-specific quantum circuits.

  • Quantum-Based Self-Attention Mechanism for Hardware-Aware Differentiable Quantum Architecture Search (2025) introduces QBSA-DQAS, a meta-learning framework that uses quantum-native self-attention and hardware-aware multi-objective optimization to automatically design parameterized quantum circuits for variational algorithms on NISQ devices. The method replaces classical similarity metrics with attention scores computed from parameterized quantum circuits, jointly optimizes noisy expressibility and Probability of Successful Trials, and then applies deterministic circuit simplifications such as commutation, fusion, elimination to shrink depth and gate count without hurting accuracy.

  • Noise-Aware Quantum Architecture Search Based on NSGA-II Algorithm (2026) A noise-aware quantum architecture search (NA-QAS) framework that designs variational quantum circuits while explicitly incorporating hardware noise during training. It combines a hybrid Hamiltonian εε-greedy strategy with a variable-depth NSGA-II algorithm to balance circuit expressibility against quantum resource overhead. Experiments on binary and Iris multi-class classification under noisy conditions show that the searched architectures achieve better performance and higher resource efficiency than existing approaches.

  • Differentiable Architecture Search for Adversarially Robust Quantum Computer Vision (2026) A hybrid quantum-classical Differentiable Quantum Architecture Search (DQAS) framework that jointly optimizes quantum circuit structure and adversarial robustness using gradient-based methods. It augments standard DQAS with a lightweight trainable Classical Noise Layer applied before the quantum circuit, allowing simultaneous learning of gate choices and noise parameters to improve robustness without sacrificing clean accuracy. Experiments on MNIST, FashionMNIST, and CIFAR, under multiple adversarial attacks and realistic quantum noise, show consistently higher clean and robust accuracy than existing quantum architecture search approaches, including on real quantum hardware.

  • Probabilistic Design of Parametrized Quantum Circuits through Local Gate Modifications (2026) Local quantum architecture search algorithm that designs parametrized quantum circuits via probabilistic, evolution-inspired gate-level modifications to existing circuits. The method restricts itself to a fixed set of local actions, enabling efficient search over architectures while adapting to task-specific requirements in quantum machine learning. Evaluations on synthetic regression tasks and two quantum chemistry datasets show that the discovered circuits achieve competitive performance and can be deployed on contemporary quantum hardware.

  • AI Agents for Variational Quantum Circuit Design (2026) An autonomous agent-based framework for designing variational quantum circuit architectures, addressing the combinatorial complexity of manual ansatz construction in quantum machine learning. The autonomous agent (AI agent) interacts with a quantum simulation environment, proposing candidate circuits, triggering fully automated training and validation, and refining its design policy from performance feedback with minimal human intervention. Experiments show that the agent can iteratively evolve simple initial ansätze into increasingly expressive architectures that improve task performance, suggesting a scalable path toward automated variational circuit development in the NISQ era.

Surveys and Reviews

Libraries and Benchmarks

Talks and Tutotials

Contributing

Please follow these steps to contribute:

  1. Fork the repo.
  2. Create a new branch (e.g., feature/new-qas).
  3. Commit your changes to the new branch.
  4. Create a Pull Request, and provide a brief description of the changes.

License

This work is licensed under a Creative Commons Attribution 4.0 International License.

About

A curated list of standout libraries, projects, tutorials, influential research papers, and essential resources focused on Quantum Architecture Search (QAS). This collection is designed to serve as a structured and thorough reference, empowering researchers and developers to accelerate their work and stay at the forefront of QAS advancements

Topics

reinforcement-learning genetic-algorithm quantum-computing evolutionary-algorithms quantum-machine-learning sample-based-learning quantum-architecture-search

Resources

Readme
Activity

Stars

13 stars

Watchers

1 watching

Forks

5 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors