Reproduce “From Characterization To Construction: Generative Quantum Circuit Synthesis from Gate Set Tomography Data” on OpenResearch

About this paper

King Yiu Yu, Aritra Sarkar, Erbing Hua, Maximilian Rimbach-Russ, Ryoichi Ishihara, Sebastian Feld

The Quantum Machine Learning Control (QMLC) framework is introduced for generative quantum circuit synthesis, directly leveraging noisy Gate Set Tomography (GST) data to generate circuits adapted to specific hardware. This approach avoids the need for ideal unitary labels or explicit noise models, offering a data-driven path to hardware-native compilation.

Key insights

• An end-to-end generative framework can directly synthesize quantum circuits from noisy empirical characterization data, bypassing traditional idealized models and costly simulations.
• Latent representations learned from aggregated noisy tomographic data can capture complex, shared physical noise environments (e.g., crosstalk, drift) crucial for hardware-aware circuit generation.
• Specialized data pre-conditioning and loss functions enable robust machine learning from partial, noisy observations, allowing the model to detect and incorporate subtle hardware error signatures without explicit noise models.

Summary by alphaXiv.