Scholarlink

<p>We introduce MNISQ, the first large-scale dataset for both quantum and classical machine learning during the NISQ era, containing 4.95 million circuits of 10 qubits constructed with up to 100 two-qubit gates. MNISQ serves as a foundational resource for developing natural language processing (NLP) models for quantum computing and deep learning models. The dataset is derived from quantum-encoded classical data (e.g., MNIST) and is available in two formats: quantum circuits and classical descriptions (Quantum Assembly Language, QASM). We perform baseline experiments on circuit classification using both quantum and classical methods. Quantum Kernel methods achieve up to 97% accuracy in multiclass classification. We also explore the impact of noise in quantum machine learning, helping develop error-mitigation strategies for noisy hardware. In classical experiments, we use QASM files with NLP models: S4, Transformer, and LSTM. The S4 model reaches 77% accuracy (81% with data augmentation), demonstrating that modern machine learning models can effectively classify quantum circuits. The dataset is publicly available at <a href="https://doi.org/10.5281/zenodo.19656638">https://doi.org/10.5281/zenodo.19656638</a> &#xa0;and related codes are available on <a href="https://github.com/FujiiLabCollaboration/MNISQ-quantum-circuit-dataset/tree/main">GitHub</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

MNISQ: A Large-Scale Quantum Circuit Dataset for Machine Learning in the NISQ Era

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月26日

查看全文
  • Leonardo Placidi,
  • Ryuichiro Hataya,
  • Toshio Mori,
  • Koki Aoyama,
  • Hayata Morisaki,
  • Kosuke Mitarai,
  • Keisuke Fujii

摘要

We introduce MNISQ, the first large-scale dataset for both quantum and classical machine learning during the NISQ era, containing 4.95 million circuits of 10 qubits constructed with up to 100 two-qubit gates. MNISQ serves as a foundational resource for developing natural language processing (NLP) models for quantum computing and deep learning models. The dataset is derived from quantum-encoded classical data (e.g., MNIST) and is available in two formats: quantum circuits and classical descriptions (Quantum Assembly Language, QASM). We perform baseline experiments on circuit classification using both quantum and classical methods. Quantum Kernel methods achieve up to 97% accuracy in multiclass classification. We also explore the impact of noise in quantum machine learning, helping develop error-mitigation strategies for noisy hardware. In classical experiments, we use QASM files with NLP models: S4, Transformer, and LSTM. The S4 model reaches 77% accuracy (81% with data augmentation), demonstrating that modern machine learning models can effectively classify quantum circuits. The dataset is publicly available at https://doi.org/10.5281/zenodo.19656638  and related codes are available on GitHub.