Quantum Reservoir Expressivity Confirmed on IBM Hardware via ORS Score

Researchers have developed a new method for evaluating quantum reservoirs, utilizing an “order-statistics (ORS) expressivity score” that streamlines the diagnostic process. Existing methods require computational resources that grow exponentially with system size; however, the ORS score assesses reservoir performance without reconstructing the full output distribution, offering an efficiency gain for scaling quantum machine learning. Importantly, the team’s framework includes a closed-form depolarizing-noise correction, making it directly usable on hardware. Validating their approach, the researchers confirmed the ORS score’s effectiveness by running tests on IBM quantum hardware, demonstrating its reliability beyond simulation and opening new avenues for characterizing these promising systems.

Quantum Reservoirs for Machine Learning Applications

A new diagnostic tool allows researchers to evaluate the performance of quantum reservoirs, promising candidates for near-term quantum machine learning, without the computational burden of analyzing their complete output, a feat previously considered impractical at scale. The method, detailed in research with a manuscript dated July 10, 2026, focuses on the most probable outcomes of a quantum reservoir ensemble, avoiding the exponential growth in resources demanded by traditional diagnostics. This efficiency gain is critical as quantum systems increase in complexity, potentially unlocking a pathway to scaling quantum machine learning evaluation.

The team, led by Laia Domingo of Centre de Visió per Computador (CVC), Universitat Autònoma de Barcelona (UAB), and Eurecat, Centre Tecnològic de Catalunya developed the ORS score with a key advantage: a depolarizing-noise correction. This correction is particularly significant because real-world quantum hardware is inherently noisy; the ability to directly apply the diagnostic to actual devices, rather than relying solely on simulations, represents a substantial step forward. “ORS is defined for arbitrary reservoir families and admits a closed-form depolarizing-noise correction, making it directly applicable to finite-shot and hardware data,” the authors explain. Validation of the ORS score extended beyond simulations, with the team successfully demonstrating its effectiveness on IBM quantum hardware. This practical verification lends considerable credibility to the findings, confirming that the score accurately reflects the intrinsic expressivity of different reservoir families.

The researchers also introduced the concept of effective rank, R_eff, to measure how much input-dependent information reaches the readout stage, complementing the ORS score and providing a more complete picture of reservoir performance. Together, these diagnostics offer a scalable and hardware-compatible framework for assessing quantum reservoirs across diverse architectures and tasks.

Order-Statistics Expressivity Score for Reservoir Diagnosis

Current methods for evaluating quantum reservoirs, essential components in emerging quantum machine learning architectures, face a fundamental scaling challenge. Existing diagnostics, designed to assess a reservoir’s ability to generate useful feature maps, demand computational resources that increase exponentially with the number of qubits. This limitation hinders progress toward larger, more complex quantum systems capable of tackling real-world problems. Researchers are now focusing on diagnostics that circumvent this exponential bottleneck, and a newly developed framework centered around the “order-statistics (ORS) expressivity score” offers a potential solution. The core innovation lies in the ORS score’s ability to evaluate reservoirs without reconstructing the full probability distribution of their outputs. Instead, the diagnostic focuses on comparing only the largest output probabilities of a reservoir ensemble against an analytical baseline derived from Haar-random states. This streamlined approach dramatically reduces computational cost, decoupling it from the Hilbert-space dimension.

Crucially, the ORS score admits a closed-form depolarizing-noise correction, making it directly usable on hardware. The researchers explain that across benchmarks, ORS captures the intrinsic expressivity hierarchy of reservoir families while R_eff determines when that expressivity translates into usable predictive information. Building on their recent work in simulation-free fidelity estimation, the team tackled the challenge of assessing reservoir quality without the exponential resource demands of existing methods. Their approach centers on two complementary quantities: the order-statistics (ORS) expressivity score and the effective rank of the feature matrix. This is particularly valuable given that “noise can even become a useful computational resource rather than only a source of degradation,” according to the study. They found that R_eff can decrease due to symmetries within the reservoir or exponential concentration of observables, highlighting the importance of considering not just the richness of the dynamics but also their usability. “Expressive reservoir dynamics improve performance only when they generate a sufficiently rich feature matrix,” the authors explain.

Beyond simply assessing whether a quantum reservoir functions, researchers are now focusing on how effectively it explores its potential computational space. A newly developed methodology centers on the “order-statistics (ORS) expressivity score,” a metric designed to evaluate quantum reservoirs without the prohibitive computational cost of reconstructing full output distributions, a significant leap forward given that existing diagnostics scale exponentially with system size. The team’s framework isn’t limited to ideal conditions; it incorporates a crucial correction for depolarizing noise, making it directly usable on hardware. This is particularly valuable considering the pervasive challenge of noise in current quantum systems, where mitigating errors is paramount. The research highlights a nuanced understanding of reservoir performance.

While quantum reservoir computing and quantum extreme learning machines promise to sidestep the optimization challenges plaguing many quantum machine learning algorithms, evaluating the quality of the underlying quantum reservoir itself has proven difficult at scale. This simplification allows the ORS score to remain cost-independent of Hilbert-space dimension, a crucial advantage as quantum systems grow in complexity. Importantly, the framework incorporates a closed-form depolarizing-noise correction, making it directly applicable to data acquired from actual quantum hardware, a significant step given the pervasive influence of noise in current devices. This enhancement provides a more robust assessment of expressivity, leveraging the basis invariance of Haar-random states. The team reports that the diagnostic’s ability to distinguish between reservoirs with genuine expressive power and those that merely appear so is noteworthy.

A single metric now accurately gauges the expressive power of quantum reservoirs, bypassing computational bottlenecks that previously hampered scaling. Researchers have moved beyond theoretical evaluations of quantum reservoirs, validating a new diagnostic tool, the order-statistics (ORS) expressivity score, against established complexity measures and, crucially, on actual quantum hardware. This advancement addresses a critical limitation of prior diagnostics, which demanded resources that grew exponentially with system size, rendering them impractical for larger, more complex quantum systems. The team led by Laia Domingo at Centre de Visió per Computador (CVC), Barcelona, demonstrated that the ORS score effectively captures the intrinsic expressivity hierarchy of different reservoir families. This efficiency is further enhanced by the fact that the ORS score admits a closed-form depolarizing-noise correction, allowing for direct application on noisy, real-world quantum devices. They found that while ORS captures the inherent potential of a reservoir, R_eff determines whether that potential translates into actual performance.

Beyond simulations, validating the ORS score required execution on actual quantum hardware. The team successfully deployed their framework on IBM quantum systems, a crucial step in establishing the practical relevance of their diagnostics. This hardware verification involved running reservoir ensembles and comparing the ORS scores obtained with those predicted by their theoretical model, confirming the score’s robustness even with the inherent noise present in current devices. The ability to apply ORS directly to hardware stems from the fact that the ORS score admits a closed-form depolarizing-noise correction, a feature that distinguishes it from many existing reservoir diagnostics. The researchers further investigated how noise impacts the ORS score’s effectiveness. This finding challenges the conventional view of noise as solely detrimental to quantum computation, suggesting that specific noise models can, in fact, enhance the expressivity of quantum reservoirs. The team demonstrated that the noise-corrected ORS gap, the difference between the ORS score of a given reservoir and that of a Haar-random baseline, remained informative even under significant simulated noise. Crucially, the study highlighted the interplay between expressivity and usability.

Existing methods for assessing reservoir quality, such as those relying on full-state reconstructions, become computationally prohibitive as system size increases, prompting the need for more efficient tools. The team’s innovation lies in the ORS score, which bypasses the need to reconstruct the entire output distribution, instead focusing on the largest output probabilities. This allows for a cost-independent analysis, unaffected by the Hilbert-space dimension, and crucially, admits a closed-form depolarizing-noise correction, making it directly usable on hardware. This combination of ORS and effective rank provides a comprehensive, scalable, and hardware-compatible diagnostic toolkit for quantum reservoirs, applicable across diverse architectures and tasks, and offering a pathway toward more robust and efficient quantum machine learning systems.