Unsupervised Method Research Articles

Over the service life of several decades, structural damage detection is crucial for ensuring the safety and durability of engineering structures. However, existing methods often overlook the spatiotemporal coupling in multi-sensor data, hindering the full exploitation of structural dynamic evolution and spatial correlations. This paper proposes an autoencoder model integrating Temporal Convolutional Networks (TCN) and Graph Attention Networks (GAT), termed TCNGAT-AE, to establish an unsupervised damage detection method. The model utilizes the TCN module to extract temporal dependencies and dynamic features from vibration signals, while leveraging the GAT module to explicitly capture the spatial topological relationships within the sensor network, thereby achieving deep fusion of spatiotemporal features. The proposed method adopts an “offline training-online detection” framework, requiring only data from the healthy state of the structure for training, and employs reconstruction error as the damage indicator. To validate the proposed method, two sets of experimentally measured data are utilized: one from the Z-24 concrete box-girder bridge under ambient excitation, and the other from the Old Ada Bridge under vehicle load excitation. Additionally, ablation studies are conducted to analyze the effectiveness of the spatiotemporal fusion mechanism. Results demonstrate that the proposed method achieves effective damage detection in both different structural types and excitation scenarios. Furthermore, the explicit modeling of spatiotemporal features significantly enhances detection performance, with the anomaly detection rate showing substantial improvement compared to baseline models utilizing only temporal or spatial modeling. Moreover, this end-to-end framework processes raw vibration signals directly, avoiding complex preprocessing. This makes it highly suitable for practical and near-real-time monitoring. The findings of this study demonstrate that the damage detection method based on TCNGAT-AE can be effectively applied to structural safety monitoring in complex engineering environments, and can be further integrated with real-time monitoring systems of critical structures for online analysis.

Pancreatic ductal adenocarcinoma (PDAC) is a rapidly progressing cancer that responds poorly to immunotherapies. Intratumoral tertiary lymphoid structures (TLS) have been associated with rare long-term PDAC survivors, but the role of TLS in PDAC and their spatial relationships within the context of the broader tumor microenvironment remain unknown. In this study, we report the generation of a spatial multiomic atlas of PDAC tumors and tumor-adjacent lymph nodes from patients treated with combination neoadjuvant immunotherapies. Using machine learning-enabled hematoxylin and eosin image classification models, imaging mass cytometry, and unsupervised gene expression matrix factorization methods for spatial transcriptomics, we characterized cellular states within and adjacent to TLS spanning distinct spatial niches and pathologic responses. Unsupervised learning identified TLS-specific spatial gene expression signatures that are significantly associated with improved survival in patients with PDAC. We identified spatial features of pathologic immune responses, including intratumoral TLS-associated B-cell maturation colocalizing with IgG dissemination and extracellular matrix remodeling. Our findings offer insights into the cellular and molecular landscape of TLS in PDACs during immunotherapy treatment.

A Level I trauma center used machine learning algorithms to identify risk factors and patterns in falls among older adults, which constitute our greatest burden of traumatic admissions. A retrospective analysis was conducted on 2,391 ground-level fall trauma admissions from 2017-2022 including variables related to demographics, and weather conditions at admission. Supervised learning models were developed to predict older adult vs younger counterpart falls. In this machine learning modality, we generated a Decision Tree, a Support Vector Machine Classifier Algorithm, and a Logistic Regression Model. Unsupervised learning methods uncover patterns or groupings in the dataset of older adult ground-level falls, which consists of 1,742 records from 2017-2022 trauma admissions including comorbidity variables. Unsupervised learning methods of Principal Components Analysis, Hierarchical Clustering, and Market Basket Analysis were employed. All three supervised models found the female sex as an important variable in predicting older adult falls. Unsupervised learning identified discernible patterns and groupings, revealing that certain weather variables are associated with falls. These machine learning modalities can shed light on what may be important risk factors for older adult falls and can help to target awareness and outreach.

Temporal action detection aims to predict temporal boundaries and category labels of actions in untrimmed videos. In the past years, many weakly supervised temporal action detection methods have been proposed to relieve the annotation cost of fully supervised methods. Due to the discrepancy between action localization and action classification, the two-branch structure is widely adopted by existing weakly supervised methods, where the classification branch is used to predict category-wise score and the localization branch is used to predict foreground score for each segment. Under the weakly supervised setting, the model training is mainly guided by the video-level or sparse segment-level annotations. As a result, the classification branch tends to focus on the most discriminative segments while ignore less discriminative ones so as to minimize the classification cost, and the localization branch may assign high foreground scores for some negative segments. This phenomenon can severely damage the action detection performance, because the foreground scores and classification scores are combined together in the testing stage for action detection. To deal with this problem, several methods have been proposed to encourage the consistency between the classification branch and localization branch. However, these methods only consider the video-level or segment-level consistency, without considering the relation among different segments to be consistent. In this paper, we propose a Cross-Task Relation-Aware Consistency (CRC) strategy for weakly supervised temporal action detection, including an intra-video consistency module and an inter-video consistency module. The intra-video consistency module can well guarantee the relationship among segments from the same video to be consistent, and the inter-video consistency module guarantees the relationship among segments from different videos to be consistent. These two modules are complementary to each other by combining both intra-video and inter-video consistency. Experimental results show that the proposed CRC strategy can consistently improve the performance of existing weakly supervised methods, including click-level supervised methods (e.g., LACP Lee et al., 2021), video-level supervised methods (e.g., DELU Chen et al., 2022) and unsupervised methods (e.g., BaS-Net Lee et al., 2020), verifying the generality and effectiveness of the proposed method.

Deep clustering of polysomnography data to characterize sleep structure in healthy sleep and non-rapid eye movement parasomnias.

Mineral prospectivity mapping for multi-source geoscience data: A novel unsupervised deep learning method

Deep learning methods have demonstrated state-of-the-art performance in image restoration, especially when trained on large-scale paired datasets. However, acquiring paired data in real-world scenarios poses a significant challenge. Unsupervised restoration approaches based on generative adversarial networks (GANs) offer a promising solution without requiring paired datasets. Yet, these GAN-based approaches struggle to surpass the performance of conventional unsupervised GAN-based frameworks without significantly modifying model structures or increasing the computational complexity. To address these issues, we propose a self-collaboration (SC) strategy for existing restoration models. This strategy utilizes information from the previous stage as feedback to guide subsequent stages, achieving significant performance improvement without increasing the framework's inference complexity. The SC strategy comprises a prompt learning (PL) module and a restorer ($Res$Res). It iteratively replaces the previous less powerful fixed restorer $\overline{Res}$Res¯ in the PL module with a more powerful $Res$Res. The enhanced PL module generates better pseudo-degraded/clean image pairs, leading to a more powerful $Res$Res for the next iteration. Our SC can significantly improve the $Res$Res 's performance by over 1.5 dB without adding extra parameters or computational complexity during inference. Meanwhile, existing self-ensemble (SE) and our SC strategies enhance the performance of pre-trained restorers from different perspectives. As SE increases computational complexity during inference, we propose a re-boosting module to the SC (Reb-SC) to improve the SC strategy further by incorporating SE into SC without increasing inference time. This approach further enhances the restorer's performance by approximately 0.3 dB. Additionally, we present a baseline framework that includes parallel generative adversarial branches with complementary "self-synthesis" and "unpaired-synthesis" constraints, ensuring the effectiveness of the training framework. Extensive experimental results on restoration tasks demonstrate that the proposed model performs favorably against existing state-of-the-art unsupervised restoration methods.

An unsupervised ensemble learning method for real-time anomaly detections in variable refrigerant flow systems

A machine learning tool to analyze spectroscopic changes in high-dimensional data.

The deployment of deep learning models in Internet of Things (IoT) systems is increasingly threatened by adversarial attacks. To address the challenge of effectively detecting adversarial examples generated by Generative Adversarial Networks (AdvGANs), this paper proposes an unsupervised detection method that integrates spatial statistical features and multidimensional distribution characteristics. First, a collection of adversarial examples under four different attack intensities was constructed on the CIFAR-10 dataset. Then, based on the VGG16 and ResNet50 classification models, a dual-module collaborative architecture was designed: Module A extracted spatial statistics from convolutional layers and constructed category prototypes to calculate similarity, while Module B extracted multidimensional statistical features and characterized distribution anomalies using the Mahalanobis distance. Experimental results showed that the proposed method achieved a maximum AUROC of 0.9937 for detecting AdvGAN attacks on ResNet50 and 0.9753 on VGG16. Furthermore, it achieved AUROC scores exceeding 0.95 against traditional attacks such as FGSM and PGD, demonstrating its cross-attack generalization capability. Cross-dataset evaluation on Fashion-MNIST confirms its robust generalization across data domains. This study presents an effective solution for unsupervised adversarial example detection, without requiring adversarial samples for training, making it suitable for a wide range of attack scenarios. These findings highlight the potential of the proposed method for enhancing the robustness of IoT systems in security-critical applications.

Rapid urbanization drives significant land use transformations, making the timely detection of newly constructed buildings a critical research focus. This study presents a novel unsupervised framework that integrates pixel-level change detection with object-level, mono-temporal building information to identify new constructions. Within this framework, we propose the Building Line Index (BLI) to capture structural characteristics from building edges. The BLI is then combined with spectral, textural, and the Morphological Building Index (MBI) to extract buildings. The fusion weight (φ) between the BLI and MBI was determined through experimental analysis to optimize performance. Experimental results on a case study in Wuhan, China, demonstrate the method’s effectiveness, achieving a pixel accuracy of 0.974, an average category accuracy of 0.836, and an Intersection over Union (IoU) of 0.515 for new buildings. Critically, at the object-level—which better reflects practical utility—the method achieved high precision of 0.942, recall of 0.881, and an F1-score of 0.91. Comparative experiments show that our approach performs favorably against existing unsupervised methods. While the single-case study design suggests the need for further validation across diverse regions, the proposed strategy offers a robust and promising unsupervised pathway for the automatic monitoring of urban expansion.

New Bayesian and deep learning spatio-temporal models can reveal anomalies in sensor data more effectively.

A new era in global health assistance requires a focus on efficiently using limited and declining donor funds. This shift requires better evaluation methods to allocate resources effectively. Most evaluations in low- and middle-income countries (LMICs) examine health disparities within countries, but it is also crucial to assess health outcomes at an inter-country level based on national wealth. Cross-country studies support resource reallocation to the neediest nations and help transition programs like HIV responses within countries with better health infrastructure. This paper presents an unsupervised machine learning method, Principal Component Analysis (PCA), applied to household surveys from 15 African countries to create a universal wealth index that allows multiple countries to be compared on a common scale. Our method places households on a regional wealth scale, enabling cross-country comparisons of health indicators. We used a pooled dataset of 136,086 households from 15- Population-based HIV Impact Assessment (PHIA) countries and validated our universal ranking approach against a local wealth indicator adjusted for macroeconomic differences. The results showed coherence between the macroeconomic-adjusted multinational scale and the PCA-created regional scale, supporting the method’s usability for regional household rankings. The proposed method relocates households, as citizens of the world, on a regional wealth scale compared to most surveys that rank them by income placements in their local states. The validation results suggest that the direction and magnitude of mobility of households from national to regional scale in both methods were adequately coherent, ensuring the usability of our approach in ranking households regionally. The PCA-created border-agnostic wealth quintiles enable policymakers to optimize their efficiency improvement efforts, which promises superior efficiency gains over the siloed localized efficiency improvements. Our approach, tested on PHIA-participating countries, can be replicated for similar surveys to study utilization patterns and health outcomes globally.

Traditional clustering algorithms are popular unsupervised methods and have been widely applied in mineral prospectivity mapping (MPM). Despite the advantages of these algorithms in terms of simplicity and popularity, they are not strong enough to struggle with high-dimensional, complex, and non-linear geospatial data. Consequently, they may lead to suboptimal clustering performance, a reason for not being able to precisely recognize and discriminate complex mineralization-related anomaly patterns in mineral exploration datasets. To improve the clustering performance, we propose a deep embedded clustering (DEC) approach for MPM. DEC is an unsupervised method that uses deep neural networks to learn from the feature representations and optimize cluster assignments simultaneously. In this study, evidence layers, representing porphyry copper mineralization, were first generated. Then, four clustering techniques were applied to generate prospectivity models. The prediction rate of the models was evaluated using the prediction-area (P-A) plot. The results showed that the prediction rates of K-means, Gaussian mixture model (GMM), DEC-K-means, and DEC-GMM prospectivity models were 66, 68, 69, and 72%, respectively. This demonstrates that DEC-based clustering outperforms conventional clustering algorithms and that DEC-GMM effectively recognizes mineralization-related patterns. Finally, to benefit from the advantages of all the applied clustering methods, we calculated a confidence index, as an ensemble technique, to recognize exploration targets, those that support further mineral exploration programs in terms of low uncertainty.

This research investigates the integration of the permissioned Blockchain Hyperledger Fabric with machine learning (ML) to develop advanced financial fraud prevention strategies. Fabric's distinctive architecture uses execute-order-validate consensus, channels, and private data collections. This creates a secure, unchangeable, and sustainable data foundation. The design ensures transactional integrity and offers a reliable source of validated data for analytical processes. This high-quality data enables ML algorithms to work effectively. Supervised and ensemble models identify known fraud, while unsupervised methods, such as anomaly detection, find new threats in real-time, surpassing traditional systems. The primary effect arises from integrating these two technologies. Fabric's secure ledger offers the protected information needed to build trustworthy ML models. As a result, these advanced models let the network discuss threats and respond to suspicious behavior immediately. This merger forms a strong security framework. Fabric guarantees data integrity. Machine learning adds smart threat identification. Together, they provide financial institutions with a more innovative and dependable shield against fraud.

Abstract Phase-only spatial light modulators (SLMs) offer versatile beam shaping, but are limited to monochromatic operation due to diffractive dispersion. A potential solution is to implement the smooth wavefront transformation of refractive freeform optics as a phase shift pattern on the SLM, called programmable freeform optics. Unfortunately, current phase-only SLMs can only inflict phase shifts of a few times 2π. This requires the use of a wrapping procedure that results in discontinuous phase jumps, reintroducing diffractive dispersion. This paper presents a differentiable design method for high-precision, shallow, refractive freeform topologies. These low-relief topologies can be implemented on an SLM, as a phase shift pattern with limited discontinuous phase jumps, allowing full-color beam shaping. This full-color performance is verified in a simulation setting, and is further experimentally analyzed in a setting with a commercial SLM. Differentiable raytracing enables the optimization of depth-constrained freeform surfaces for various source-target settings, but is limited by its computational speed. To address this limitation, an unsupervised learning method is furthermore introduced that predicts freeform topologies with a point source. By wrapping the predicted, resulting phase shifts, this approach enables full-color, real-time beam shaping with a single SLM.

Metal implants and other high-density objects cause significant artifacts in computed tomography (CT) images, hindering clinical diagnosis. Traditional metal artifact reduction methods often leave residual artifacts due to sinogram edges discontinuities. Supervised deep learning approaches struggle due to reliance on paired data, while unsupervised methods often lack multi-domain information. In this paper, we propose TDMAR-Net, a diffusion model-based three-domain neural network that leverages priors from projection, image, and Fourier domains for removing metal artifact and enhancing CT image quality. To enhance the model's learning capability and gradient optimization while preventing reliance on a single data structure, we employ a two-stage training strategy that combines large-scale pretraining with masked data fine-tuning, improving both accuracy and adaptability in metal artifact removal. The specific process is to adjust the weight of the high frequency and low frequency components of the input image through the high-pass filter module in the Fourier domain, and process the image into blocks to extract the diffusion prior information. The prior information is then introduced iteratively into the sinogram and image domains to fill in the metal-induced artifacts. Our method overcomes the challenges of information sharing and complementarity across different domains, ensuring that each domain contributes effectively, thereby enhancing the precision and robustness of metal artifact elimination. Experiments show that our approach superior to existing unsupervised methods, which we have validated on both synthetic and clinical datasets.

This research presents unsupervised machine learning and statistical methods to identify and analyze tidal tails in open star clusters using data from the Gaia DR3 catalog. We aim to identify member stars and to detect and analyze tidal tails in five open clusters, BH 164, Alessi 2, NGC 2281, NGC 2354, and M67, of ages between 60 Myr and 4 Gyr. These clusters were selected based on the previous evidence of extended tidal structures. We utilized machine learning algorithms such as Density-based Spatial Clustering of Applications with Noise (DBSCAN) and principal component analysis (PCA), along with statistical methods to analyze the kinematic, photometric, and astrometric properties of stars. Key characteristics of tidal tails, including radial velocity, the color-magnitude diagram, and spatial projections in the tangent plane beyond the cluster's Jacobi radius (r_J), were used to detect them. We used N-body simulations to visualize and compare the observables with real data. Further analysis was done on the detected cluster and tail stars to study their internal dynamics and populations, including the binary fraction. We also applied the residual velocity method to detect rotational patterns in the clusters and their tails. We identified tidal tails in all five clusters, with detected tails extending farther in some clusters and containing significantly more stars than previously reported (tails ranging from 40 to 100 pc, one to four times their r_J, with 100-200 tail stars). The luminosity functions of the tails and their parent clusters were generally consistent, and tails lacked massive stars. In general, the binary fraction was found to be higher in the tidal tails. Significant rotation was detected in M67 and NGC 2281 for the first time.

Mass spectrometry imaging (MSI) enables label-free molecular mapping in tissues but presents challenges for spatial segmentation due to high dimensionality, nonlinear spectral variation, and tissue heterogeneity. Traditional unsupervised clustering methods often rely on predefined cluster numbers and overlook spatial information, yielding fragmented or biologically implausible results. We introduce MSInet, a self-supervised deep learning framework for robust, annotation-free MSI segmentation. MSInet combines two strategies within a convolutional neural network: patch-wise contrastive learning to capture global semantic relationships, and superpixel-guided refinement to enforce local spatial consistency. This dual-consistency design simultaneously enhances global context awareness and local boundary precision during training. MSInet was evaluated on MALDI-MSI of mouse brain, DESI-MSI of renal tumor, and a synthetic data set with ground truth. It consistently outperformed state-of-the-art methods (e.g., t-SNE + k-means, CNNAE + region-growing, and GCN-based models), achieving higher accuracy and biological fidelity. On simulated data, MSInet achieved an Adjusted Rand Index of 0.89 and Normalized Mutual Information of 0.86, with ∼25.8% ARI improvement over baselines. It also precisely delineated complex anatomical subregions in the brain (Silhouette Coefficient = 0.78) and distinguished tumor, necrosis, and healthy regions in renal tissues, closely aligning with histological references. MSInet further demonstrated robustness to MSI noise. By integrating global and local contextual modeling in a self-supervised architecture, MSInet offers a powerful, scalable solution for accurate and biologically meaningful MSI segmentation, with broad potential for spatial omics and biomedical applications.

The ongoing challenge of doping in sports has triggered the adoption of advanced scientific strategies for the detection and prevention of doping abuse. This review examines the potential of integrating metabolomics aided by artificial intelligence (AI) and machine learning (ML) for profiling small-molecule metabolites across biological systems to advance anti-doping efforts. While traditional targeted detection methods serve a primarily forensic role—providing legally defensible evidence by directly identifying prohibited substances—metabolomics offers complementary insights by revealing both exogenous compounds and endogenous physiological alterations that may persist beyond direct drug detection windows, rather than serving as an alternative to routine forensic testing. High-throughput platforms such as UHPLC-HRMS and NMR, coupled with targeted and untargeted metabolomic workflows, can provide comprehensive datasets that help discriminate between doped and clean athlete profiles. However, the complexity and dimensionality of these datasets necessitate sophisticated computational tools. ML algorithms, including supervised models like XGBoost and multi-layer perceptrons, and unsupervised methods such as clustering and dimensionality reduction, enable robust pattern recognition, classification, and anomaly detection. These approaches enhance both the sensitivity and specificity of diagnostic screening and optimize resource allocation. Case studies illustrate the value of integrating metabolomics and ML—for example, detecting recombinant human erythropoietin (r-HuEPO) use via indirect blood markers and uncovering testosterone and corticosteroid abuse with extended detection windows. Future progress will rely on interdisciplinary collaboration, open-access data infrastructure, and continuous methodological innovation to fully realize the complementary role of these technologies in supporting fair play and athlete well-being.