Check the expressive power of quantum reservoir on IBM hardware with ORS score

Researchers have developed a new method to assess quantum reservoirs using an order statistics (ORS) expressiveness score that streamlines the diagnostic process. Existing methods require computational resources that grow exponentially with system size. However, the ORS score evaluates reservoir performance without reconstructing the complete output distribution, providing increased efficiency for scaling quantum machine learning. Importantly, the team’s framework includes a closed-form depolarization noise correction that can be used directly on the hardware. To validate their approach, the researchers confirmed the validity of the ORS score by running tests on IBM quantum hardware, demonstrating reliability beyond simulation and opening new avenues for characterizing these promising systems.

Quantum reservoir for machine learning applications

A new diagnostic tool will allow researchers to assess the performance of quantum reservoirs, which are promising candidates for short-term quantum machine learning, without incurring the computational burden of analyzing the complete output, which was previously considered impractical at scale. The method, detailed in a manuscript study dated July 10, 2026, focuses on the most likely outcomes of quantum reservoir ensembles and avoids the exponential increase in resources required for traditional diagnostics. As the complexity of quantum systems increases, this increase in efficiency is critical and could pave the way for scaling quantum machine learning evaluations.

A team led by Laia Domingo from the Center de Visió per Computador (CVC), Universitat Autònoma de Barcelona (UAB), and Eurecat, Center Tecnològic de Catalunya has developed an ORS score with the important advantage of depolarization noise correction. This correction is especially important because real-world quantum hardware is inherently noisy. The ability to apply diagnostics directly to real devices, rather than relying solely on simulation, represents a significant advance. “ORS is defined for arbitrary reservoir families and allows closed-form depolarization noise correction, so it can be applied directly to finite-shot or hardware data,” the authors explain. Validation of the ORS score extended beyond simulation, and the team was able to demonstrate its effectiveness on IBM quantum hardware. This practical validation lends considerable confidence to the results and ensures that the scores accurately reflect the unique expressiveness of different reservoir families.

The researchers also introduced the concept of effective rank R._Huh?which measures how much input-dependent information reaches the readout stage, complements the ORS score and provides a more complete picture of the reservoir’s performance. Together, these diagnostics provide a scalable and hardware-compatible framework for evaluating quantum reservoirs across a variety of architectures and tasks.

Order statistical expressiveness scores for reservoir diagnostics

Current methods for evaluating quantum reservoirs, an essential component of new quantum machine learning architectures, face fundamental scaling challenges. Existing diagnostics designed to assess a reservoir’s ability to generate useful feature maps require computational resources that increase exponentially with the number of qubits. This limitation hinders progress toward larger and more complex quantum systems that can tackle real-world problems. Researchers are currently focusing on diagnostics that circumvent this exponential bottleneck, and a newly developed framework centered around the Ordinary Statistics (ORS) Expressiveness Score provides a potential solution. The core of the innovation lies in the ability of the ORS score to evaluate a reservoir without reconstructing the complete probability distribution of the reservoir’s yield. Instead, this diagnostic focuses on comparing only the maximum output probability of the reservoir ensemble to an analytical baseline derived from Haar random conditions. This streamlined approach significantly reduces computational costs and decouples them from the Hilbert spatial dimension.

Importantly, the ORS score admits closed-form depolarization noise correction, making it directly available in the hardware. The researchers found that ORS captures the unique representation hierarchy of reservoir families across benchmarks, while R_Huh? Determine when that expressiveness is translated into usable predictive information. Building on recent work in simulation-free fidelity estimation, the team took on the challenge of assessing reservoir quality without the exponential resource demands of existing methods. Their approach focuses on two complementary quantities: the order statistics (ORS) expressiveness score and the effective rank of the feature matrix. This is especially valuable given that, according to the study, “noise is not just a source of degradation and can even be a useful computational resource.” They have R_Huh? It can be reduced by symmetries within the reservoir and exponential concentrations of observables, highlighting the importance of considering not only the richness of the dynamics but also their usefulness. “Representative reservoir dynamics improve performance only when generating a sufficiently rich feature matrix,” the authors explain.

Researchers are now focused on how to effectively explore the computational potential of quantum reservoirs, rather than just assessing whether they work. The newly developed methodology centers around the Order Statistics (ORS) expressiveness score, a metric designed to assess quantum reservoirs without the prohibitive computational cost of reconstructing the complete output distribution. This is a significant advance considering that existing diagnostics scale exponentially with system size. Team frameworks are not limited to ideal situations. Important corrections for depolarization noise are included and made available directly in the hardware. This is particularly valuable given the pervasive challenges of noise in current quantum systems, where error mitigation is paramount. This research focuses on a nuanced understanding of reservoir performance.

Quantum reservoir computing and quantum limit learning machines promise to avoid the optimization challenges that plague many quantum machine learning algorithms, but assessing the quality of the underlying quantum reservoir itself has proven difficult at scale. This simplification allows the ORS score to remain independent of the cost of Hilbert spatial dimensions, which is an important advantage as quantum systems become more complex. Importantly, this framework incorporates closed-form depolarization noise correction, making it directly applicable to data acquired from real quantum hardware, an important step given the pervasive effects of noise in current devices. This enhancement leverages the basis invariance of Haar random states to provide a more robust evaluation of expressiveness. The researchers report that the diagnostic’s ability to distinguish between truly expressive reservoirs and those that merely appear to be is noteworthy.

A single metric will now be able to accurately measure the expressive power of a quantum reservoir, avoiding computational bottlenecks that previously hindered scaling. Beyond a theoretical assessment of quantum reservoirs, the researchers validated a new diagnostic tool, the Ordinal Statistics (ORS) expressiveness score, against established complexity measures and, importantly, on actual quantum hardware. This advance overcomes a significant limitation of previous diagnostics, where resources increased rapidly with system size, making them impractical for larger, more complex quantum systems. A team led by Laia Domingo from the Center de Visió per Computador (CVC) in Barcelona has demonstrated that ORS scores effectively capture the inherent expressiveness hierarchy of different reservoir families. This efficiency is further enhanced by the fact that the ORS score allows closed-form depolarization noise correction and can be directly applied to noisy real-world quantum devices. They found that while ORS captures the inherent potential of the reservoir, R_Huh? Determine whether that potential is reflected in actual performance.

Beyond simulation, validating the ORS score required execution on real quantum hardware. The team successfully deployed the framework on an IBM quantum system, an important step in establishing the practical relevance of the diagnostics. This hardware validation involved running the reservoir ensemble and comparing the resulting ORS scores to those predicted by the theoretical model, confirming the robustness of the scores despite the inherent noise present in current devices. The ability to apply ORS directly to the hardware stems from the fact that the ORS score allows closed-form depolarization noise correction, which distinguishes it from many existing reservoir diagnostics. The researchers further investigated how noise affects the validity of ORS scores. This finding challenges the traditional view that noise is only detrimental to quantum computation and suggests that certain noise models can indeed enhance the expressive power of quantum reservoirs. The research team demonstrated that the noise-corrected ORS gap, the difference between the ORS score of a given reservoir and the Haar random baseline ORS score, remains informative even under significant simulated noise. Importantly, this study focused on the interplay between expressiveness and ease of use.

Existing methods for assessing reservoir quality, such as those that rely on full-state reconstructions, become computationally prohibitive as system size increases, increasing the need for more efficient tools. The team’s innovation lies in the ORS score, which eliminates the need to reconstruct the entire output distribution and instead focuses on the maximum output probability. This enables a cost-independent analysis that is insensitive to Hilbert spatial dimensions and, more importantly, enables closed-form depolarization noise correction, which can be used directly in hardware. This combination of ORS and Effective Rank provides a comprehensive, scalable, and hardware-compatible diagnostic toolkit for quantum reservoirs that can be applied to a variety of architectures and tasks, providing a path toward more robust and efficient quantum machine learning systems.