Search Results (10,992)

Unsupervised Hebbian learning is a biologically inspired algorithm designed to extract representations from input images, which can subsequently support supervised learning. It presents a promising alternative to traditional artificial neural networks (ANNs). Many attempts have focused on enhancing Hebbian learning by incorporating more biologically plausible components. Contrarily, we draw inspiration from recent advances in ANNs to rethink and further improve Hebbian learning in three interconnected aspects. First, we investigate the issue of overfitting in Hebbian learning and emphasize the importance of selecting an optimal number of training epochs, even in unsupervised settings. In addition, we discuss the risks and benefits of anti-Hebbian learning in model performance, and our visualizations reveal that synapses resembling the input images sometimes do not necessarily reflect effective learning. Then, we explore the impact of different activation functions on Hebbian representations, highlighting the benefits of properly utilizing negative values. Furthermore, motivated by the success of large pre-trained language models, we propose a novel approach for leveraging unlabeled data from other datasets. Unlike conventional pre-training in ANNs, experimental results demonstrate that merging trained synapses from different datasets leads to improved performance. Overall, our findings offer fresh perspectives on enhancing the future design of Hebbian learning algorithms. Full article

Amid ongoing efforts to develop extremely large, multimodal models, there is increasing interest in efficient Small Language Models (SLMs) that can operate without reliance on large data-centre infrastructure. However, recent SLMs (e.g., LLaMA or Phi) with up to three billion parameters are predominantly trained in high-resource languages, such as English, which limits their applicability to industries that require robust NLP solutions for less-represented languages and low-resource settings, particularly those requiring low latency and adaptability to evolving label spaces. This paper examines a retrieval-based approach to multi-label text classification (MLC) for a media monitoring dataset, with a particular focus on less-represented languages, such as Slovene. This dataset presents an extreme MLC challenge, with instances labelled using up to twelve thousand categories. The proposed method, which combines retrieval with computationally efficient prediction, effectively addresses challenges related to multilinguality, resource constraints, and frequent label changes. We adopt a model-agnostic approach that does not rely on a specific model architecture or language selection. Our results demonstrate that techniques from the extreme multi-label text classification (XMC) domain outperform traditional Transformer-based encoder models, particularly in handling dynamic label spaces without requiring continuous fine-tuning. Additionally, we highlight the effectiveness of this approach in scenarios involving rare labels, where baseline models struggle with generalisation. Full article

Gastrointestinal disorders, such as colorectal cancer (CRC), pose a substantial health burden worldwide, showing increased incidence rates across different age groups. Detecting and removing polyps promptly, recognized as CRC precursors, are crucial for prevention. While traditional colonoscopy works well, it is vulnerable to specialist errors. This study suggests an AI-based diminutive sessile polyp localization assistant utilizing the YOLO-V8 family. Comprehensive evaluations were conducted using a diverse dataset that was assembled from various available datasets to support our investigation. The final dataset contains images obtained using two imaging methods: white light endoscopy (WLE) and narrow-band imaging (NBI). The research demonstrated remarkable results, boasting a precision of 96.4%, recall of 93.89%, and F1-score of 94.46%. This success can be credited to a meticulously balanced combination of hyperparameters and the specific attributes of the comprehensive dataset designed for the colorectal polyp localization in colonoscopy images. Also, it was proved that the dataset selection was acceptable by analyzing the polyp sizes and their coordinates using a special matrix. This study brings forth significant insights for augmenting the detection of diminutive sessile colorectal polyps, thereby advancing technology-driven colorectal cancer diagnosis in offline scenarios. This is particularly beneficial for gastroenterologists analyzing endoscopy capsule images to detect gastrointestinal polyps. Full article

We present a systematic machine learning study of the solubility of diverse pharmaceutical acids in deep eutectic solvents (DESs). Using an automated Dual-Objective Optimization with Iterative feature pruning (DOO-IT) framework, we analyze a solubility dataset compiled from the literature for ten pharmaceutically important carboxylic acids and augment it with new measurements for mefenamic and niflumic acids in choline chloride- and menthol-based DESs, yielding N = 1020 data points. The data-driven multi-criterion measure is applied for final model selection among all collected accurate and parsimonious models. This three-step procedure enables extensive exploration of the model’s hyperspace and effective selection of models fulfilling notable accuracy, simplicity, and also persistency of the descriptors selected during model development. The dual-solution landscape clarifies the trade-off between complexity and cost in QSPR for DES systems and shows that physically meaningful energetic descriptors can replace or enhance explicit COSMO-RS predictions depending on the application. Full article

In this study, subject-independent (inter-subject), multiple-session electroencephalography (EEG) data classification was tested for loving-kindness meditation (LKM) and non-meditation. This is a novel study that extends our previous work on intra-subject, multiple-session classification. Here, two meditation techniques, LKM-Self and LKM-Other, were independently compared with non-meditation. For each mental task, five sessions of data collected from each of the twelve participants were placed in a common data pool, from which randomly selected session data were used for training and testing the machine learning algorithms. Three previously tested BCI pipelines were used. In each case, feature extraction was performed using common spatial patterns (CSPs), short-time Fourier transform (STFT), or a fusion of CSP and STFT, followed by classification using a neural network structure. This study was further divided into three cases, where two, three, or four session pairs were used to train the algorithms, and the remaining session pair was used for testing. For each individual instance, the test was repeated thirty times to generalize the results. Thus, a total of 9900 independent tests were conducted for the entire dataset. The mean classification accuracies obtained in this study were lower than those reported in our previous intra-subject classification study. For example, in LKM-Self/non-meditation classification using three session pairs with the CSP + STFT pipeline, the mean accuracy for all tests was 62.3%, with the bottom 50% at 46.0% and the top 50% at 78.3%, demonstrating variability across session selections. The corresponding intra-subject classification result for the same instance was 72.1%. For all other instances, a similar pattern was observed. Furthermore, when considering all mean accuracies obtained, in 83.3% of the instances, CSP + STFT produced better classification accuracies than CSP or STFT alone. At the same time, in 75.0% of the instances, increasing the number of training session pairs led to improved classification accuracies. This study demonstrates that the subject-independent, multiple-session EEG classification of meditation and non-meditation is feasible for specific session combinations. Further research is needed to confirm these findings across larger and more diverse participant groups. These findings provide a foundation for developing subject-independent algorithms that can guide long-term meditation practice. Full article

The purpose of this study was to determine the motivation to buy eco-friendly products via machine learning techniques. With this in mind, a dataset was collected between November and December 2024 from 245 organic consumers in Maramureș County, Romania, via a questionnaire. Consumers’ main motivations to buy eco-friendly products were considered according to three categories: Health Care, Environmental Protection, and Superior Quality. In the analysis of the dataset, among the four feature selection techniques used, Random Forest was determined to be the best with the highest accuracy value. At the beginning of the study, 16 variables were thought to be important categorical factors for consumers’ eco-friendly product-buying motivations, with 5 of these being found to be the most effective with the Random Forest technique. Then, the SHAP method was applied to identify the contribution of driving factors to the buying motivation for eco-friendly products. All analyses were conducted with Python software. The results of the SHAP method indicated that while all factors perform well, consumers considering themselves as eco-friendly is the most important factor for the Environmental Protection category when buying eco-friendly products, while the most important criterion of the original certification category was found to be the Health Care category. The most effective factor for Superior Quality was determined as the high-price category, which is the main barrier to purchasing eco-friendly products. Full article

As the maritime industry transitions toward Maritime Autonomous Surface Ships (MASS), Vessel Traffic Service Operators (VTSOs) face new challenges in managing mixed traffic of conventional and autonomous vessels. Effective VTSO selection is becoming increasingly critical for maritime safety, yet current recruitment processes rely on subjective methods that limit objective evaluation of candidate suitability. This study presents the first machine learning-based classification model for VTSO recruitment. Eight features were defined, including sea service experience, navigation career, education, certifications, and language proficiency. Due to limited access to actual recruitment data, expert-validated simulated datasets were constructed through labeling by 40 maritime professionals and density estimation-based augmentation. Four algorithms were compared, with XGBoost achieving 94.6% F1-score. Feature importance analysis revealed TOEIC score as the most critical predictor, followed by seafaring career, with 3–4 years of experience identified as optimal. These findings indicate that English proficiency for communication with shore remote control centers and practical maritime experience for assessing autonomous vessel behaviors constitute core VTSO competencies in the MASS era. The proposed model demonstrates potential to improve subjective recruitment methods by discovering quantifiable competency patterns, offering a pathway toward data-driven, standardized, and transparent decision-making for enhanced maritime safety. Full article

Background: Recurrent cervical cancer is one of the most defining threats to patient longevity, underscoring the need for prognostic models to identify high-risk patients. Objectives: The aim of the study is to integrate clinical data with the GSE44001 Dataset to identify key risk factors associated with the recurrence of cervical cancer. Patients are stratified into high-, moderate-, and low-risk groups using selected clinical and molecular features. Identifying a long non-coding RNA (lncRNA) gene signature associated with recurrent cervical cancer. Methods: From the total data collected, 138 recurrent cervical cancer patients were identified. GSE44001 Dataset is downloaded from the NCBI GEO Database. When using the GENCODE Annotation tool, the long non-coding RNA is filtered. The dataset is then linked with filtered long non-coding RNA. The Least Absolute Shrinkage Selection Operator (LASSO) is employed to find attributes in gene expression analysis. Risk factors of recurrent cervical cancer are identified. Risk value is assigned to each individual based on the selected lncRNAs and the corresponding overfitting coefficients. Result: The RNN Long Short-Term Memory model demonstrates a prognostic value, where high-risk patients experience a shorter duration of recurrence-free survival (p < 0.05). Individuals with a recurrence of cervical carcinoma, a progressive disease, were associated with the ATXN8OS marker, the C5orf60 indicator, and the INE1 index gene. In contrast, patients diagnosed at earlier stages are aligned with the KCNQ1DN marker, LOH12CR2 gauge, RFPL1S value, and KCNQ1OT1 indicator. Patients in moderate stages were primarily associated with the EMX2OS score. Conclusions: The research findings demonstrate that the nine-lncRNA signature, when combined with deep learning, offers a powerful approach for recurrence risk stratification in cervical cancer. Full article

The use of Artificial Intelligence (AI) in healthcare has significantly advanced early disease detection, enabling timely diagnosis and improved patient outcomes. This work proposes an end-to-end machine learning (ML) model for predicting diabetes based on data quality by following key steps, including advanced preprocessing by KNN imputation, intelligent feature selection, class imbalance with a hybrid approach of SMOTEENN, and multi-model classification. We rigorously compared nine ML classifiers, namely ensemble approaches (Random Forest, CatBoost, XGBoost), Support Vector Machines (SVM), and Logistic Regression (LR) for the prediction of diabetes disease. We evaluated performance on specificity, accuracy, recall, precision, and F1-score to assess generalizability and robustness. We employed SHapley Additive exPlanations (SHAP) for explainability, ranking, and identifying the most influential clinical risk factors. SHAP analysis identified glucose levels as the dominant predictor, followed by BMI and age, providing clinically interpretable risk factors that align with established medical knowledge. Results indicate that ensemble models have the highest performance among the others, and CatBoost performed the best, which achieved an ROC-AUC of 0.972, an accuracy of 0.968, and an F1-score of 0.971. The model was successfully validated on two larger datasets (CDC BRFSS and a 130-hospital dataset), confirming its generalizability. This data-driven design provides a reproducible platform for applying useful and interpretable ML models in clinical practice as a primary application for future Internet-of-Things-based smart healthcare systems. Full article

Human Pose Estimation (HPE) models have varied applications and represent a cutting-edge branch of study, whose systems such as MediaPipe (MP), OpenPose (OP), and AlphaPose (ALP) show marked success. One of these areas, however, that is inadequately researched is the impact of image degradation on the accuracy of HPE models. Image degradation refers to images whose visual quality has been purposefully degraded by means of techniques, such as brightness adjustments (which can lead to an increase or a decrease in the intensity levels), geometric rotations, or resolution downscaling. The study of how these types of degradation impact the performance functionality of HPE models is an under-researched domaina that is a virtually unexplored area. In addition, current methods of the efficacy of existing image restoration techniques have not been rigorously evaluated and improving degraded images to a high quality has not been well examined in relation to improving HPE models. In this study, we explicitly clearly demonstrate a decline in the precision of the HPE model when image quality is degraded. Our qualitative and quantitative measurements identify a wide difference in performance in identifying landmarks as images undergo changes in brightness, rotation, or reductions in resolution. Additionally, we have tested a variety of existing image enhancement methods in an attempt to enhance their capability in restoring low-quality images, hence supporting improved functionality of HPE. Interestingly, for rotated images, using Pillow of OpenCV improves landmark recognition precision drastically, nearly restoring it to levels we see in high-quality images. In instances of brightness variation and in low-quality images, however, existing methods of enhancement fail to yield the improvements anticipated, highlighting a large direction of study that warrants further investigation and calls for additional research. In this regard, we proposed a wide-ranging system for classifying different types of image degradation systematically and for selecting appropriate algorithms for image restoration, in an effort to restore image quality. A key finding is that in a related study of current methods, the Tuned RotNet model achieves 92.04% accuracy, significantly outperforming the baseline model and surpassing the official RotNet model in predicting rotation degree of images, where the accuracy of official RotNet and Tuned RotNet classifiers were 61.59% and 92.04%, respectively. Furthermore, in an effort to facilitate future research and make it easier for other studies, we provide a new dataset of reference images and corresponding degenerated images, addressing a notable gap in controlled comparative studies, since currently there is a lack of controlled comparatives. Full article

Background/Objectives: Lung cancer (NSCLC), which accounts for approximately 85% of lung cancers, exhibits marked heterogeneity that complicates molecular characterization and treatment selection. Recent advances in deep learning (DL) have enabled the extraction of genomic-related morphological features directly from routine Hematoxylin and Eosin (H&E) histopathology, offering a potential complement to Next-Generation Sequencing (NGS) for precision oncology. This review aimed to evaluate how DL models have been applied to predict molecular alterations in NSCLC using H&E-stained slides. Methods: A systematic search following PRISMA 2020 guidelines was conducted across PubMed, Scopus, and Web of Science to identify studies published up to March 2025 that used DL models for mutation prediction in NSCLC. Eligible studies were screened, and data on model architectures, datasets, and performance metrics were extracted. Results: Sixteen studies met inclusion criteria. Most employed convolutional neural networks trained on publicly available datasets such as The Cancer Genome Atlas (TCGA) to infer key mutations including EGFR, KRAS, and TP53. Reported areas under the curve ranged from 0.65 to 0.95, demonstrating variable but promising predictive capability. Conclusions: DL-based histopathology shows strong potential for linking tissue morphology to molecular signatures in NSCLC. However, methodological heterogeneity, small sample sizes, and limited external validation constrain reproducibility and generalizability. Standardized protocols, larger multicenter cohorts, and transparent validation are needed before these models can be translated into clinical practice. Full article

Body measurement plays a crucial role in cattle breeding selection. Traditional manual measurement of cattle body size is both time-consuming and labor-intensive. Current automatic body measurement methods require expensive equipment, involve complex operations, and impose high computational costs, which hinder efficient measurement and broad application. To overcome these limitations, this study proposes an efficient automatic method for cattle body measurement. Lateral and dorsal image datasets were constructed by capturing cattle keypoints characterized by symmetry and relatively fixed positions. A lightweight SCW-YOLO keypoint detection model was designed to identify keypoints in both lateral and dorsal cattle images. Building on the detected keypoints, 11 body measurements—including body height, chest depth, abdominal depth, chest width, abdominal width, sacral height, croup length, diagonal body length, cannon circumference, chest girth, and abdominal girth—were computed automatically using established formulas. Experiments were performed on lateral and dorsal datasets from 61 cattle. The results demonstrated that the proposed method achieved an average relative error of 4.7%. Compared with the original model, the parameter count decreased by 58.2%, compute cost dropped by 68.8%, and model size was reduced by 57%, thus significantly improving lightweight efficiency while preserving acceptable accuracy. Full article

With the widespread deployment of high-voltage and ultra-high-voltage transmission lines, composite insulators play a vital role in modern power systems. However, prolonged service leads to material aging, and the current lack of standardized, quantitative methods for evaluating silicone rubber degradation poses significant challenges for condition-based maintenance. To address this measurement gap, we propose a novel aging assessment framework that integrates Fourier Transform Infrared (FTIR) spectroscopy with a measurement-oriented ensemble learning model. FTIR is utilized to extract absorbance peak areas from multiple aging-sensitive functional groups, forming the basis for quantitative evaluation. This work establishes a measurement-driven framework for aging assessment, supported by information-theoretic feature selection to enhance spectral relevance. The dataset is augmented to 4847 samples using linear interpolation to improve generalization. The proposed model employs k-nearest neighbor (KNN), Support Vector Machine (SVM), Random Forest (RF), and Gradient-Boosting Decision Tree (GBDT) within a two-tier ensemble architecture featuring dynamic weight allocation and a class-balanced weighted cross-entropy loss. The model achieves 96.17% accuracy and demonstrates strong robustness under noise and anomaly disturbances. SHAP analysis confirms the resistance to overfitting. This work provides a scalable and reliable method for assessing silicone rubber aging, contributing to the development of intelligent, data-driven diagnostic tools for electrical insulation systems. Full article

Crack segmentation in cluttered scenes with slender and irregular patterns remains difficult, and practical systems must balance accuracy and efficiency. We present CONTI-CrackNet, which is a lightweight visual state-space network that integrates a Multi-Directional Selective Scanning Strategy (MD3S). MD3S performs bidirectional scanning along the horizontal, vertical, and diagonal directions, and it fuses the complementary paths with a Bidirectional Gated Fusion (BiGF) module to strengthen global continuity. To preserve fine details while completing global texture, we propose a Dual-Branch Pixel-Level Global–Local Fusion (DBPGL) module that incorporates a Pixel-Adaptive Pooling (PAP) mechanism to dynamically weight max-pooled responses and average-pooled responses. Evaluated on two public benchmarks, the proposed method achieves an F1 score (F1) of 0.8332 and a mean Intersection over Union (mIoU) of 0.8436 on the TUT dataset, and it achieves an mIoU of 0.7760 on the CRACK500 dataset, surpassing competitive Convolutional Neural Network (CNN), Transformer, and Mamba baselines. With 512 × 512 input, the model requires 24.22 G floating point operations (GFLOPs), 6.01 M parameters (Params), and operates at 42 frames per second (FPS) on an RTX 3090 GPU, delivering a favorable accuracy–efficiency balance. These results show that CONTI-CrackNet improves continuity and edge recovery for thin cracks while keeping computational cost low, and it is lightweight in terms of parameter count and computational cost. Full article

Breath analysis is a non-invasive diagnostic method that offers insights into both physiological and pathological conditions. Exhaled breath contains volatile organic compounds, which act as biomarkers for disease detection, allowing for the monitoring of treatments and the tailoring of medicine to individuals. Recent advancements in chemical sensing, mass spectrometry, and spectroscopy have improved the ability to identify these biomarkers; however, traditional statistical approaches often struggle to handle the complexities of breath data. Artificial intelligence (AI) and machine learning (ML) have revolutionized breath analysis by uncovering intricate patterns among volatile breath markers, enhancing diagnostic precision, and facilitating real-time disease identification. Despite significant progress, challenges remain, including issues with data standardization, model interpretability, and the necessity for extensive and varied datasets. This study reviews the applications of ML in analyzing breath volatile organic compounds, highlighting methodological shortcomings and obstacles to clinical validation. A thorough literature review was performed using the PubMed and Scopus databases, which included studies that focused specifically on the role of machine learning in disease diagnosis and incidence prediction via breath analysis. Among the 524 articles reviewed, 97 satisfied the specified inclusion criteria. The selected studies applied ML techniques, fell within the scope of this review, and emphasize the potential of ML models for non-invasive diagnostics. The findings indicate that traditional ML methods dominate, while ensemble methods are on the rise, and deep learning (DL) techniques (especially CNNs and LSTMs) are increasingly used for classifying respiratory diseases. Techniques for feature selection (such as PCA and ML-based methods) were frequently implemented, though challenges related to explainability and data standardization persist. Future studies should focus on enhancing model transparency and developing methods to further integrate AI into the clinical setting to facilitate early disease detection and advance precision medicine. Full article