Can a handful of atoms outperform a much larger digital neural network on a real-world task? The answer may be yes. In a study published in Physical Review Letters, a team led by Prof. PENG Xinhua and Assoc. Prof. LI Zhaokai from the University of Science and Technology of China of the Chinese Academy of Sciences demonstrated that a quantum processor comprising just nine interacting spins outperformed classical networks with thousands of nodes in realistic weather forecasting tasks.

By exploiting unique quantum features such as superposition and entanglement, quantum devices offer new ways to represent and process information. Recent experiments have shown their advantages on specialized benchmark tasks, but extending these gains to real-world applications remains a challenge. In particular, many quantum approaches rely on complex circuits that are difficult to implement accurately on today’s noisy hardware.

In this study, the researchers realized that the natural dynamics of quantum systems could inherently provide rich computational power, bypassing the need for deep quantum circuits. This led to reservoir computing, a brain-inspired machine learning approach in which a dynamical system processes incoming signals and retains memory on its own, without precise control.

In the implementation, input signals were encoded into quantum states whose entangled evolution naturally processed information in ways that are difficult to simulate classically. Even dissipation, which was usually seen as harmful in quantum computing, was turned into a useful resource for regulating the system's memory. By harnessing these native dynamics rather than fighting against them, this approach was naturally better suited to near-term quantum devices.

Using nuclear magnetic resonance techniques, the researchers built a quantum reservoir computer based on nine interacting atomic spins. They first tested it on a widely used time-series prediction benchmark known as NARMA, which achieved the best performance reported among experimental quantum approaches, reducing prediction errors by one to two orders of magnitude compared with previous circuit-based implementations.

The researchers then tested it in weather forecasting which is vital but difficult. Experimental results showed that the quantum model accurately captured temperature trends over several days. Besides, the researchers compared it with a standard classical reservoir model known as the echo state network. Notably, the nine-spin quantum reservoir achieved higher accuracy than classical reservoir networks with thousands of nodes in multi-day forecasts.

The findings provide experimental evidence that a quantum machine-learning system may outperform much larger classical counterparts on realistic tasks. This work suggests that harnessing the native dynamics of current quantum devices rather than waiting for fully fault-tolerant quantum computers may promote useful applications.