anomaly detection

Graph machine learning often suffers from distributional shift between training and test data in real-world scenarios, undermining model robustness. This paper systematically investigates the Graph Out-of-Distribution (GOOD) detection problem—identifying whether test graphs deviate from the training distribution. We formally define GOOD detection for the first time and propose a unified taxonomy encompassing four methodological categories: augmentation-based, reconstruction-based, message-passing-based, and classification-based approaches. Crucially, we rigorously distinguish GOOD detection from related tasks—including graph anomaly detection, outlier detection, and OOD generalization—highlighting its unique objectives and assumptions. Furthermore, we establish the first comprehensive theoretical framework and methodological spectrum for GOOD detection, and publicly release Awesome-GOOD-Detection, an authoritative open-source repository. This resource provides standardized benchmarks, reproducible evaluation tools, and a roadmap for future research, thereby laying foundational infrastructure for the emerging field of GOOD detection.

Scientific machine learning experiments often suffer from distorted performance evaluations due to poor experimental design and inconsistent documentation. To address this, we propose a principled framework for ML experimentation tailored to scientific research, encompassing data preprocessing, model selection, cross-validation, and reporting—emphasizing reproducibility, fair comparison, and transparency. Our key contributions include two novel quantitative metrics: the Logarithmic Overfitting Ratio (LOR) and Composite Overfitting Score (COS), which jointly characterize overfitting severity and instability across cross-validation folds. Complementing these, we introduce standardized preprocessing protocols, rigorously defined strong baselines, and modular visualization templates for diagnostic analysis. Empirical evaluation demonstrates that our framework substantially enhances experimental rigor, reproducibility, and result credibility in scientific ML. It further enables robust performance assessment and cross-study comparability, providing systematic support for establishing reliable benchmarks.

Misuse of statistical hypothesis tests severely undermines scientific reliability. This paper proposes a formal verification methodology for statistical programs: preconditions—such as normality, independence, and homoscedasticity—are explicitly encoded as logical assertions in source code; static verification is then performed on OCaml implementations using the Why3 platform to automatically detect missing or conflicting assumptions. The approach innovatively integrates contract-based programming with formal verification, distinguishing between formalizable preconditions (amenable to automated checking) and non-formalizable ones (requiring expert judgment), thereby establishing a human-in-the-loop verification paradigm. Evaluated on canonical statistical tests—including Student’s *t*-test and ANOVA—the method successfully identifies widespread misuses, such as applying the *t*-test to non-normal data or neglecting homoscedasticity checks. Results demonstrate significant improvements in the correctness, auditability, and reproducibility of statistical software.

Risk valuation systems are susceptible to undetected errors caused by data failures, misconfigurations, or anomalies, potentially leading to significant operational losses. This work proposes EQAF, a hierarchical unsupervised ensemble framework for anomaly detection that uniquely integrates domain-specific deterministic rules with multiple complementary statistical outlier detection methods to enable real-time integrity monitoring of risk computation outputs. EQAF effectively identifies subtle anomalies—such as “frozen values”—that are often missed by conventional purely statistical approaches. Experimental evaluation on four real-world risk datasets demonstrates that EQAF achieves F1 scores between 61% and 79% and improves AUC-ROC by 4–6 percentage points over the best individual baseline method, substantiating its robustness and effectiveness.

Outlier detection (OD) has long been hindered by the absence of anomaly labels under unsupervised settings. To address this, we propose DOUST—the first unsupervised OD method to incorporate test-time training (TTT), enabling online model adaptation solely using unlabeled test samples. DOUST integrates self-supervised pretraining, reconstruction loss, and consistency regularization to dynamically refine the model during inference, without requiring any anomaly annotations. Theoretical analysis shows that, given a moderately sized test set, DOUST asymptotically approaches the performance upper bound of fully supervised OD. Extensive experiments on multiple standard benchmarks demonstrate that DOUST significantly outperforms existing unsupervised methods; notably, with larger test sets, its AUC nears the supervised upper bound—providing the first empirical validation of strong generalization capability for pure test-time learning in outlier detection.

Detecting out-of-distribution (OOD) samples at test time remains challenging: ID-only methods suffer from limited discriminative capacity, while leveraging external anomaly data introduces privacy risks and task misalignment. Method: We propose AUTO, the first framework for *test-time adaptive OOD detection*, which requires no predefined anomaly data. Instead, it dynamically leverages unlabeled, real-world OOD samples from the incoming test stream to continuously refine the detector online. Contributions/Results: AUTO introduces three key components: (i) an in-out-aware filter for safe in-distribution sample selection; (ii) a dynamic memory module enabling robust replay of historical OOD patterns; and (iii) a prediction alignment objective preserving model stability. Guided by pseudo-labels, online gradient calibration, and test-time model adaptation, AUTO significantly outperforms state-of-the-art methods across standard, multi-OOD, and temporal OOD benchmarks—achieving superior detection accuracy and generalization robustness.

This work addresses the common practice of treating out-of-distribution (OOD) detection and in-distribution (ID) misclassification prediction as separate tasks, despite their intrinsic connection in building reliable classifiers. To bridge this gap, the authors propose SURE+, a unified framework that jointly models both tasks through a dual-scoring mechanism. The study further introduces novel joint evaluation metrics—DS-F1 and DS-AURC—to holistically assess performance across OOD and ID failure detection. Comprehensive experiments on the OpenOOD benchmark demonstrate that SURE+ significantly outperforms conventional single-score approaches, with particularly pronounced gains in scenarios involving easy or far-OOD samples. This work thus establishes a new paradigm and benchmark for trustworthy classification by explicitly integrating OOD detection and ID error prediction into a cohesive framework.

This work addresses the performance degradation of machine learning interatomic potentials caused by noise from unconverged or inconsistent electronic structure calculations in training data. Existing denoising approaches rely on manual curation or iterative retraining, which are computationally expensive. To overcome this limitation, the authors propose an unsupervised online denoising method that dynamically tracks the loss distribution during a single training run using exponential moving averages, enabling real-time detection and automatic down-weighting of anomalous samples without requiring additional reference calculations or iterative retraining. The method successfully recovers accurate diffusion coefficients from unconverged liquid water data and reduces energy prediction errors by a factor of three on the SPICE dataset of organic molecules, demonstrating high efficiency, scalability, and robustness.

Current evaluation practices for supervised learning models are often misleading due to an overreliance on single aggregate metrics, which neglect the alignment among data characteristics, task objectives, and real-world application contexts. This work reframes model evaluation as a context-dependent, decision-oriented process and systematically investigates—through controlled experiments—the impact of dataset properties, validation strategies, class imbalance, and asymmetric error costs on evaluation outcomes. Leveraging diverse benchmark datasets, multiple validation protocols, and multidimensional performance measures, the study uncovers common pitfalls such as the accuracy paradox, data leakage, and metric misuse. It proposes a structured evaluation framework explicitly aligned with operational goals, offering principled guidance for developing more robust, reliable, and trustworthy supervised learning systems.

This study addresses the lack of rigorous statistical assessment for the reliability of output structures in complex clustering pipelines that involve multiple data-dependent stages such as anomaly detection, feature selection, and clustering. To bridge this gap, the work systematically applies selective inference to the entire clustering analysis workflow, establishing a statistical framework that enables valid significance testing of final cluster assignments. The proposed method rigorously controls the type I error rate at any pre-specified nominal level and demonstrates strong empirical performance on both synthetic and real-world datasets. By doing so, it provides a principled and reliable foundation for statistical inference in multi-stage, data-driven clustering procedures.

This work addresses the challenge of root cause identification in high-dimensional, nonlinear causal systems with uncertainty by proposing a score-based attribution method. The approach estimates the score function—the gradient of the log-likelihood—and integrates gradients along the path from an anomalous point back to the normal data distribution to accumulate feature contributions. To the best of our knowledge, this is the first method to directly leverage the score function for root cause attribution, satisfying multiple axioms of Shapley values as well as an asymmetry axiom derived from the underlying causal structure. The method is both scalable and uncertainty-aware. Experimental results on synthetic graph data and real-world datasets from cloud services and supply chains demonstrate superior attribution accuracy and computational efficiency compared to state-of-the-art baselines.