The work, led by Shahreen Sultana, Mutasim Fuad Sarker, Md Adnan Arefeen and Sumaiya Tabassum Nimi of North South University in Dhaka, together with Md Faiyaz Bin Younus of the University of Missouri-Kansas City, is presented under the name Q-SCOPE, a framework for characterizing quantum state geometry to support out-of-distribution prediction and benchmark evaluation. The researchers point out that while the machine learning community has developed an extensive arsenal of out-of-distribution detection techniques for classical neural networks, none of these methods had been rigorously studied in the context of hybrid classical-quantum models. That omission matters because these hybrid models are increasingly being proposed for real-world deployment, and a classifier that cannot distinguish familiar data from unfamiliar data is a classifier that cannot be trusted in safety-critical settings.

The core problem the team set out to solve is the poor separation between in-distribution and out-of-distribution samples in the classification output space. Classical neural network classifiers assign probabilities through a softmax layer, and it has long been known that these confidence scores are poorly calibrated: a network trained exclusively on handwritten digits may assign high confidence to a photograph of a truck simply because the machinery that produced the output was never designed to express the concept of “not a digit.” Hybrid classical-quantum neural network models inherit this weakness, since their final classification stage is likewise a classical layer interpreting outputs from a parameterized quantum circuit. The danger is not merely theoretical; in domains such as automated driving perception, where deep learning safety concerns have been documented in the recent literature, failure to detect novel inputs can translate directly into physical risk.

To establish a baseline, the researchers first benchmarked existing out-of-distribution detection approaches drawn from the classical neural network literature, applying them to hybrid classical-quantum classifiers using standard datasets and standard evaluation metrics. This benchmarking exercise revealed the limitations of transplanting classical techniques into the quantum-enhanced setting. Methods that rely on the statistics of classical logits or on softmax confidence scores do not take advantage of the distinctive representational structure that quantum circuits provide, and their performance on hybrid models proved inconsistent across benchmark pairs. The authors argue that this is precisely where a quantum-native approach should have the advantage, because the intermediate representations produced by a quantum circuit are not arbitrary vectors but genuine quantum states living in a Hilbert space, with all the geometric structure that implies.

The novel strategy proposed in the paper exploits exactly that structure. Rather than treating the outputs of the quantum layer as opaque feature vectors, the researchers identify subspaces within the quantum feature space based on the categorical label information of the training data. Each class of the classification problem is associated with a characteristic region of the quantum state space, and the geometry of these regions becomes the foundation for detection. The key quantity is the fidelity score, a measure drawn from the quantum computing literature that quantifies the similarity between two quantum states. In quantum mechanics, fidelity ranges over the overlap between states, taking values close to one for nearly identical states and approaching zero for orthogonal ones. It has long served as the standard yardstick for comparing quantum states in tasks ranging from quantum state tomography to the characterization of quantum operations, and the Q-SCOPE framework now repurposes it as a similarity metric for machine learning.

The detection procedure works as follows. For each class in the training set, the model characterizes the typical fidelity relationships between samples that belong to that class, effectively learning a class-wise boundary in fidelity space defined by the geometry of the quantum states. When a new sample arrives, its quantum state representation is compared against these class-wise regions using the fidelity score. If the maximum fidelity with any known class region falls within the learned boundaries, the sample is treated as in-distribution; if it falls outside, the sample is flagged as out-of-distribution. The authors emphasize that this class-wise characterization is essential, because the multi-cluster structure of quantum-enhanced feature space means that a single global threshold would blur together distributions that are meaningfully distinct at the level of individual classes. Alongside the fidelity-based scoring, the framework also evaluates alternative scoring functions, including mean-based approaches, Gaussian mixture models and one-class support vector machines, under both class-wise and global formulations, providing a comprehensive comparison of how different assumptions about the feature space geometry affect detection quality.

The empirical evaluation spans multiple standard benchmark datasets. The researchers trained hybrid classical-quantum classifiers on image datasets including MNIST and Fashion-MNIST and then tested detection against a range of out-of-distribution counterparts, including Omniglot, CIFAR-10 and a comprehensive Bangla handwritten character database. These pairings were chosen to span a spectrum of difficulty, from relatively easy cases such as natural color images versus grayscale digits, to genuinely challenging cases where the in-distribution and out-of-distribution data share visual characteristics. Across these benchmarks, the team demonstrated empirically that the fidelity score, combined with the class-wise boundary characterization, successfully separates in-distribution from out-of-distribution samples, substantiating the central claim of the paper. The accompanying analysis includes extensive density plot visualizations, covering every scoring method and every in-distribution and out-of-distribution combination evaluated, which document the statistical separation achieved in fidelity space and illustrate concretely how class merging under a global formulation degrades detection performance.

The significance of the result extends beyond the immediate technical contribution. Quantum machine learning has matured from a theoretical curiosity into a field with demonstrated applications in anomaly detection across diverse domains, including high-energy physics, network security, industrial control systems and time series analysis. Quantum autoencoders, quantum kernel methods and variational quantum circuits have all been proposed for detecting unusual or anomalous data. But the out-of-distribution detection problem addressed by Q-SCOPE is subtly different from anomaly detection: rather than identifying rare events within a known domain, it asks whether a classifier can recognize when its own assumptions no longer hold. This question is foundational to the trustworthy deployment of any learning system, and until now it had not been systematically examined for the hybrid quantum models that dominate near-term quantum machine learning research conducted on noisy intermediate-scale quantum devices.

The study also carries implications for how quantum machine learning models should be benchmarked. By explicitly framing out-of-distribution detection as a benchmark evaluation task and providing a standardized comparison of classical methods against the new fidelity-based approach, the researchers give the community a template for assessing not just accuracy but reliability. As the authors note, understanding quantum machine learning fully requires rethinking generalization itself, and the behavior of models on data outside their training distribution is a central component of generalization. A hybrid quantum classifier that achieves high accuracy on test data drawn from the same distribution as its training set may still be unfit for deployment if it cannot signal when the world deviates from that distribution. Fidelity, in this light, functions as a bridge between two literatures: it is the natural similarity measure of quantum information theory, and it now serves as the detection statistic that makes quantum state geometry actionable for machine learning reliability.

The authors report that the code used in the study will be made available upon reasonable request, and they note that the research received no specific grant from funding agencies in the public, commercial or not-for-profit sectors. The work was conducted at North South University’s Department of Electrical and Computer Engineering and at the University of Missouri-Kansas City’s Division of Computing, Analytics and Mathematics, with Sumaiya Tabassum Nimi, who originated the research idea and supervised the project, serving as corresponding author. Sultana and Sarker conceived the study, conducted the experiments and drafted the main manuscript, while Younus contributed to the literature review and Arefeen supported technical and editorial revisions.

Looking ahead, the Q-SCOPE framework opens several avenues for follow-up research. The class-wise fidelity boundaries learned by the current method are defined by the geometry of specific trained models and datasets, and extending the approach to larger quantum circuits, higher-dimensional datasets and hardware implementations on real quantum processors remains an open challenge. The interplay between noise in quantum hardware and the stability of fidelity-based detection thresholds is another natural question, since noisy quantum states perturb the very geometric structure the method depends on. Nevertheless, the demonstration that a metric as fundamental as quantum fidelity can outperform transplanted classical heuristics on hybrid models marks a meaningful step toward quantum machine learning systems that know the limits of their own knowledge. In a field where enthusiasm often outpaces reliability engineering, a framework that quantifies exactly how far a new sample sits from the quantum states the model truly understands offers something rare: a principled measure of a quantum model’s self-awareness.

The task is known as out-of-distribution detection, or OOD detection: deciding, at test time, whether a new input departs from the data used to train the model. It has become one of the central safety problems in modern machine learning, because the distributional shifts that appear after deployment — a hospital adopting a new assay, a sensor drifting out of calibration, a patient presenting with an atypical disease pattern — are rarely known in advance, so a detector must be learned from in-distribution data alone. Most of the OOD literature has grown up around image and text classification, where class probabilities, logits, and deep learned representations supply convenient raw material for anomaly scores. Tabular problems with continuous or time-to-event outcomes, such as predicting blood pressure or survival time from structured clinical measurements, lack that class-based architecture. Existing tools tend to fail in one of two ways: they ignore the fitted model entirely and treat every statistical rarity as an alarm, or they rely on predictive uncertainty that is computed globally and cannot distinguish a dangerous shift in a critical biomarker from a harmless oddity in an irrelevant variable.

The new method, called OutPro — short for OOD using variable priority — rests on a deceptively simple principle: whether a data point is unusual should be judged relative to the prediction task, not relative to the full covariate distribution. The authors formalize this through the idea of a predictive subspace, the subset of input variables through which the outcome’s conditional distribution actually depends on the inputs. A case can look wildly atypical along nuisance coordinates while remaining perfectly well supported for prediction, and conversely a localized shift in a handful of influential variables, or in the dependence among them, can be nearly invisible in the full space yet devastating for the forecast. OutPro is model-aware, meaning the trained model itself enters the score: change the training outcomes and the score changes. It is subspace-aware, meaning the final comparison is confined to coordinates the model found informative. Both properties emerge from a single learned object: the decision rules of a supervised random forest, each rule a chain of simple conditions, such as age above sixty and tumor length below three centimeters, that carves a rectangular region out of the input space.

The machinery works in two connected steps. First, the forest’s rules are mined for variables carrying predictive information through variable priority, a technique the authors developed in earlier work: within each rule region, a variable’s constraint is released — deleted while every other bound stays fixed — and the variable earns high priority when releasing it consistently changes the outcome behavior of the training points inside the region. This yields a signal set of selected coordinates together with importance weights. Second, the same release operation builds a reference neighborhood for each test input. Every forest rule containing the test point is relaxed one selected variable at a time, and the researchers count how often each training case reappears in these expanded release regions. Two training points equidistant from the test case in ordinary Euclidean distance can score very differently: one may co-occur with the test point across all released coordinates, while the other appears only after relaxing a couple. A proximity score combining the total appearance count with the Gini impurity of the co-occurrence profile favors cases that appear frequently and evenly across predictive variables. The highest-scoring cases form the neighborhood, and the OOD score is the average distance to them, computed only on selected coordinates and weighted by priority.

Benchmarking began in a controlled laboratory. The team simulated data from the classic Friedman regression model, a twenty-feature setup in which only five variables actually influence the response, then perturbed test points with additive shifts ranging from a whisper — five percent of a standard deviation — to a shout of two full standard deviations. The experiments surfaced a subtlety the field has largely overlooked: a shifted point is not necessarily anomalous at all. Because the simulated covariates are independent and bounded, a nudged point often lands entirely within the original support and represents a perfectly legitimate input; only shifts that push at least one coordinate outside its observed range are truly out of distribution, and the researchers labeled ground truth accordingly. Across one hundred replications scored by the area under the precision-recall curve, the OutPro product score achieved the best average rank at every shift magnitude, with a Manhattan-distance variant close behind. Classical tabular detectors — Isolation Forest, one-class support vector machines, the local outlier factor, robust nearest-neighbor density — closed the gap only as shifts grew large enough to manufacture obviously low-density points, precisely the easy regime.

Real data demanded a more versatile adversary. The researchers built an anomaly generator from copula theory, the branch of statistics that separates marginal distributions from dependence structure. A latent Gaussian vector encodes dependence, a probability integral transform places every coordinate on a common uniform scale, and inverse marginal distributions map the result back to the observed data space. Perturbing different stages of this three-stage pipeline produces three distinct anomaly modes: warp, which distorts the tail behavior of individual marginals; joint, which relocates points to atypical regions of the dependence structure while preserving the marginals; and support, which pushes points beyond the observed range of the data altogether. Applied to sixty-one regression datasets drawn from the Penn Machine Learning Benchmark, spanning ten to 124 features and sample sizes from 47 to just over a thousand, the results split cleanly along mode lines. Under warp, density-driven classics such as one-class SVMs, Isolation Forest, and the local outlier factor led the field. Under joint and support — the modes that tangle dependence or escape the support — OutPro procedures dominated, confirming that local, prediction-derived profiles beat generic full-space scoring rules exactly where those rules are blind.

High-dimensional biology provided a sterner test. Five microarray survival studies — diffuse large B-cell lymphoma, breast cancer, lung adenocarcinoma, acute myeloid leukemia, and mantle cell lymphoma — carry feature counts that dwarf their sample sizes. The authors converted survival outcomes into continuous pseudo-responses using out-of-bag mortality predictions from random survival forests and filtered genes by Cox-score ranking before generating copula anomalies in all three modes. The pattern repeated: OutPro variants led the joint mode outright and occupied four of the six top spots under support, while Isolation Forest and other density methods retained their edge under warp. The team then dismantled the method piece by piece to verify that each component earns its place. Stripping away the variable-priority weights hurt performance in every mode; dissolving the subspace restriction back to all coordinates diluted the signal; and replacing the forest-derived neighborhood with an ordinary nearest-neighbor search in the full covariate space performed poorly everywhere. Sensitivity analyses showed the method tolerates its main tuning choice, the neighborhood size — set by default to as much as a tenth of the training sample, far larger than conventional nearest-neighbor methods — with only moderate gains from enlarging it, and total runtimes stayed under thirty seconds throughout.

The clinical payoff came from the Worldwide Esophageal Cancer Collaboration, a multi-institution registry of patients treated with esophagectomy alone for esophageal cancer. The team analyzed 6,142 adenocarcinoma cases described by 35 variables, focusing on pT3 and pT4 tumors that have invaded deeply into or through the esophageal wall. Surgery for such patients involves lymphadenectomy, the removal of lymph nodes to stage disease and strip away involved tissue, but the right number to remove has long been debated. The researchers constructed thirty-one train-test scenarios, holding out node-positive patients whose removed-node count met or exceeded a cutoff that climbed from zero to thirty. As the cutoff rose, mean OOD percentile scores fell: patients subjected to more extensive lymphadenectomy looked progressively less anomalous relative to the remaining cohort, mirroring the survival gains visible in the underlying data. At a 95th percentile threshold, roughly three removed nodes sufficed for patients with one-to-two or three-to-six positive nodes, but about thirty nodes were required when seven or more nodes were involved. Because a surgeon cannot know nodal status during the operation, the analysis supports removing on the order of thirty nodes whenever a deeply invasive tumor is suspected — squarely consistent with an earlier estimate from the same collaboration of 29 to 50 nodes depending on histopathologic type.

The method has boundaries the authors state plainly. OutPro reads only covariates at test time, so a pure concept shift — one that leaves the input distribution unchanged but alters the outcome relationship — leaves no observable trace, a limitation shared by every input-based detector. No single subspace distance dominated across all settings, and the score inherits whatever errors creep into subspace estimation. Yet the framework’s virtues are considerable: it requires no outcome labels for the test cases it scores, generalizes across regression, classification, and survival settings through its random forest backbone, runs in seconds even on genomic-scale problems, and ships as the open-source R package varPro on CRAN, developed with support from the National Institutes of Health. Beyond the operating room, the approach speaks to any field where a model’s silent ignorance carries a price, from credit risk to industrial monitoring. What the study ultimately offers is a shift of perspective: anomalousness, in prediction, is not a property of a data point alone but a relationship between a point and the task a model was trained to perform. OutPro makes that relationship measurable, one relaxed rule at a time.