image: This illustration draws a parallel between quantum state tomography and natural language modeling. In quantum tomography, structured measurements yield probability outcomes that are aggregated to reconstruct a quantum state. Similarly, in a language model, characters are selected to form words which are finally composed a sentence. This analogy highlights how both processes rely on structured input to infer complex underlying information.
Credit: ©Science China Press
An international research team has published a comprehensive review on how machine learning techniques are advancing the estimation and control of quantum systems. The study, led by Prof. Daoyi Dong from the Australian Artificial Intelligence Institute at the University of Technology Sydney, Australia, and Dr. Bo Qi from the State Key Laboratory of Mathematical Sciences, Chinese Academy of Sciences, China, offers new insights into how these data-driven methods can help tackle some of the key challenges in building practical and scalable quantum technologies.
As quantum computing, simulation, and sensing continue progressing, precise control and accurate characterization of quantum systems are becoming increasingly important. Traditional techniques often face limitations due to noise, system complexity, and limited access to system models. The authors highlight how machine learning offers adaptive, data-driven alternatives that can enhance both the robustness and efficiency of quantum operations.
The review covers a wide range of machine learning methods used in quantum estimation tasks, including approaches for reconstructing the states or dynamics of quantum systems from measurement data. Tools such as neural networks, generative models, and attention-based architectures like Transformers have shown promise for quantum tomography. One particularly interesting idea is the analogy between language modeling and quantum estimation—suggesting that reconstructing a quantum state from structured measurements is akin to assembling a sentence from characters and words.
In terms of quantum control, the paper outlines how learning-based methods can optimize control strategies under realistic constraints. Gradient-based techniques are shown to improve control fidelity and robustness when integrated with different data-driven techniques. Evolutionary algorithms are highlighted for their effectiveness in optimizing quantum systems without the need for explicit physical models. In particular, the authors introduced experimental examples involving femtosecond laser pulses, where the algorithms optimize selective control of molecular fragmentation with enhanced robustness against parameter fluctuations.
The review further explores reinforcement learning approaches that enable autonomous control through trial-and-error interactions with quantum systems. The model-free and adaptive nature makes it particularly useful for handling complex scenarios involving unknown system dynamics or partial observability. A key focus is quantum error correction, a fundamental requirement for achieving fault-tolerant quantum computing. The authors highlight recent progress in applying reinforcement learning to adaptive quantum error correction, where agents learn to select quantum gates or measurements based on real-time feedback.
By bridging AI with quantum engineering, this review outlines a promising direction toward intelligent quantum systems that are both scalable and resilient. It serves as a timely resource for researchers aiming to integrate machine learning into the design, estimation, and control of next-generation quantum devices.
See the article: Machine Learning for Estimation and Control of Quantum Systems at https://doi.org/10.1093/nsr/nwaf269
Journal
National Science Review