Curriculum-based TRADES for Robust Skin Lesion Classification under Adversarial Attacks

Seismic data interpolation and denoising play vital roles in obtaining complete and clean data in seismic data processing. Seismic data usually misses along various spatial axes and always mix with random noise. In order to obtain complete and clean seismic data, reconstruction technology can interpolate missing data and attenuate random noise. Nonsubsampled contourlet transform is an effective transform to obtain multi-scale and multi-direction sparse domain data for compression sensing interpolation and denoising. However, conventional iterative shrinkage/thresholding cannot handle ill-posed and ill-conditioned equations for solving linear inverse problem. We present a two-step iterative log thresholding method to overcome ill-posed and ill-conditioned problems and improve the convergence rate and solution accuracy, which can interpolate and denoise seismic data simultaneously in the nonsubsampled contourlet transform framework. First, we use nonsubsampled contourlet transform to convert the seismic missing data with random noise to sparse domain. Then, we apply two-step iterative log thresholding algorithm to interpolate and denoise data in sparse domain. The result of each iteration is based on the results of the previous two iterations, which can accelerate convergence rate. In addition, log thresholding can further improve convergence rate and solution accuracy. Finally, we use inverse nonsubsampled contourlet transform to obtain the interpolated and denoised seismic data. The new method can reconstruct the irregularly missing data and attenuate random noise to obtain complete and clean seismic data with high accuracy, which is crucial for seismic imaging and inversion. We demonstrate the applicability and effectiveness of this simultaneous interpolation and denoising technique with successful applications to both synthetic and field data examples.

The rise of real-time payment systems such as India’s Unified Payments Interface (UPI) has transformed transaction behavior while introducing new security challenges. Al-though machine learning is widely used for fraud detection, most prior work studies adversarial robustness in experimental settings that do not reflect operational constraints of real payment systems. We propose a framework that incorporates UPI-specific limitations — fixed transaction attributes and rate caps — into adversarial robustness evaluation. Under realistic, constraint-aware attacks, we show that standard models (e.g., XGBoost) can be severely degraded (F1 falling from 0.71 to 0.04). SHAP-based analysis reveals how adversarial manipulations distort model reasoning. To defend against such attacks, we introduce UPI-Counterfactual Invariance Regularization (UPI-CIR), a training regularizer that enforces prediction invariance to constrained perturbations. Compared with conventional adversarial training, UPI-CIR substantially improves survivability under attack (F1 = 0.82) and also yields better performance on clean data. Our work establishes a more realistic standard for adversarial testing in financial systems and proposes a practical, domain-aware defense for securing modern payment infrastructures.

Dataset distillation (DD) allows datasets to be distilled to fractions of their original size while preserving the rich distributional information so that models trained on the distilled datasets can achieve a comparable accuracy while saving significant computational loads. Recent research in this area has been focusing on improving the accuracy of models trained on distilled datasets. In this paper, we aim to explore a new perspective of DD. We study how to embed adversarial robustness in distilled datasets, so that models trained on these datasets maintain the high accuracy and meanwhile acquire better adversarial robustness. We propose a new method that achieves this goal by incorporating curvature regularization into the distillation process with much less computational overhead than standard adversarial training. Extensive empirical experiments suggest that our method not only outperforms standard adversarial training on both accuracy and robustness with less computation overhead but is also capable of generating robust distilled datasets that can withstand various adversarial attacks.

Background and AimIn 2015, the Japanese Society of Gastroenterological Surgery (JSGS) initiated data verification in the gastroenterological section of the National Clinical Database (NCD) and reported high accuracy of data entry. Remote audits were introduced for data validation on a trial basis in 2016 and formally accepted into use in 2017‐2018. The aim of this study was to audit the data quality of the NCD gastroenterological section for 2016‐2018 and to confirm the high accuracy of data in remote audits.MethodsEach year, 45‐46 hospitals were selected for audit. Twenty cases were randomly selected in each hospital, and the accuracy of patient demographic and surgical outcome data (46 items) was compared with the corresponding medical records obtained by visiting the hospital (site‐visit audit) or by mailing data from the hospital to the JSGS office (remote audit).ResultsA total of 136 hospitals were included, of which 88 (64.7%) had a remote audit, and 124 936 items were evaluated with an overall data accuracy of 98.1%. There was no significant difference in terms of data accuracy between site‐visit audit and remote audit. Accuracy, sensitivity, and specificity of mortality were 99.7%, 89.7%, 100% for site‐visit audits and 99.8%, 97.3%, 100% for remote audits, respectively. Mean time spent on data verification per case of remote audits was shorter than that of site‐visit audits (10.0 minutes vs 13.7 minutes, P < 0.001).ConclusionThe audits showed that NCD data are reliable and characterized by high accuracy. Remote audits may substitute site‐visit audits.

Phishing attacks are becoming increasingly sophisticated, underscoring the need for detection systems that strike a balance between high accuracy and computational efficiency. This paper presents a comparative evaluation of traditional Machine Learning (ML), Deep Learning (DL), and quantized small-parameter Large Language Models (LLMs) for phishing detection. Through experiments on a curated dataset, we show that while LLMs currently underperform compared to ML and DL methods in terms of raw accuracy, they exhibit strong potential for identifying subtle, context-based phishing cues. We also investigate the impact of zero-shot and few-shot prompting strategies, revealing that LLM-rephrased emails can significantly degrade the performance of both ML and LLM-based detectors. Our benchmarking highlights that models like DeepSeek R1 Distill Qwen 14B (Q8_0) achieve competitive accuracy, above 80%, using only 17 GB of VRAM, supporting their viability for cost-efficient deployment. We further assess the models’ adversarial robustness and cost-performance tradeoffs, and demonstrate how lightweight LLMs can provide concise, interpretable explanations to support real-time decision-making. These findings position optimized LLMs as promising components in phishing defence systems and offer a path forward for integrating explainable, efficient AI into modern cybersecurity frameworks.

Bias in medical image segmentation can lead to unequal performance across demographic subgroups, raising concerns about fairness and reliability in clinical AI systems. While deep learning models have achieved high segmentation accuracy, ensuring equitable performance across race and gender remains a significant challenge, particularly in privacy-sensitive healthcare environments. This study investigates fairness-aware medical image segmentation for hip and knee radiographs using deep learning models evaluated in both centralized and Federated Learning (FL) settings. We introduce Curriculum Learning (CL) strategies and Progressive Loss (PL) functions to regulate sample difficulty during training. In addition, we propose two novel fairness-oriented federated learning algorithms, Federated Intersection over Union (FedIoU) and Federated Intersection over Union with Outlier Analysis (FedIoUoutlier). Experiments are conducted using multiple segmentation backbones and simulated multi-site data partitions derived from the Osteoarthritis Initiative dataset. Model performance is evaluated using Intersection over Union (IoU), IoU standard deviation, Skewed Error Ratio (SER), and Min-Max Disparity across race and gender subgroups. Statistical significance was verified using paired t-tests to compare per-sample IoU performance against baseline configurations. Across both hip and knee segmentation tasks, curriculum learning and progressive loss strategies consistently improved segmentation accuracy and reduced demographic performance disparities in centralized training. In federated settings, fairness-aware aggregation further enhanced performance. Notably, FedIoUoutlier combined with balanced curriculum learning and tiered progressive loss achieved the highest mean IoU while yielding the lowest SER and Min-Max Disparity, indicating improved fairness without sacrificing accuracy. In several configurations, federated models matched or exceeded the performance of optimized centralized models, with statistically significant improvements in per-sample IoU over baseline configurations. The results demonstrate that structured training strategies and fairness-aware federated aggregation can jointly improve accuracy, stability, and demographic fairness in medical image segmentation. By integrating curriculum learning, progressive loss, and novel FL algorithms, this work provides a practical pathway toward equitable and privacy-preserving AI systems for medical imaging.

Data inaccuracy is an important problem in wireless sensor networks, since the accuracy is affected by harsh environments and malicious nodes. The reason for this data inaccuracy is the improper identification of outliers. To detect exact outliers in the wireless sensor networks, we propose the relative correlation based clustering (RCC) technique with high data accuracy and low computational overhead. Identifying spatial, temporal correlation and attribute correlation is the first phase of the proposed algorithm. The second phase is optimal cluster formation and outlier classification based on two correlation levels. The inference of the proposed idea shows high outlier detection rate with different outlier corruption level. Moreover, our results when compared with previous approach taking the same data into consideration clearly outperform them, identifying high level of detection rate (99.87%) in the top-line with near to the ground false alarm rate.

The constant demand for food and beverages to sustain human life drives fierce competition among manufacturers, focusing on product excellence in terms of timeliness, quality, and pricing. The key to competitiveness depends in optimizing manufacturing processes by efficiently utilizing company resources. To ensure the overall optimization and reliable flow of manufacturing processes, a systematic evaluation process must be used, Overall Equipment Efficiency (OEE) stands out as a prominent performance measurement metric in manufacturing process efficiency. OEE serves as a valuable diagnostic tool, exposing areas for improvement and losses transparently. Accurate OEE measurement necessitates the implementation of an automated data collection system with minimum human dependencies, human intervention, and conducting on-the-fly calculations to informed the stakeholder/user. Data quality and accuracy in OEE measurement is very critical. Low quality and accuracy data could lead to false decision. OEE categorizes losses into six groups loss to pinpoint significant factors for potential improvement. Once OEE could be maintain at high level with high data accuracy and right improvement point, an optimum manufacturing process, and cost effective in manufacturing expenses will be achieve. Base on the result comparison for OEE result before and after the system implementation, positive improvement in OEE could reach 8.06%. This scenario be adopted by other company, and could become a model for 1st phase journey in company digital transformation.

We propose RobustQuote, a novel defense framework designed to enhance the adversarial robustness of vision transformers. The core idea is to leverage trusted reference images drawn from a dynamically changing pool unknown to the attacker as contextual anchors to detect and correct adversarial perturbations in input images. By quoting semantic features from uncorrupted references, RobustQuote mitigates the propagation of corrupted features through the model and uses these references to compute the rectification term. RobustQuote consists of two key modules: a quotation mechanism that propagates the global semantic tokens (i.e., the [cls] tokens) from reference images, and a rectification mechanism that adjusts the image tokens of the adversarial input using contextual signals from those references. During training, the model is explicitly guided to detect and rectify adversarial inputs more aggressively than clean ones. This approach is modular and plug-and-play, making it compatible with a wide range of vision transformer architectures, such as DeiT. Experiments show that RobustQuote achieves adversarial robustness on par with TRADES-trained DeiT under strong threat models, even when the attacker is aware of the reference set. In a more realistic setting where the attacker lacks access to the references, RobustQuote outperforms the second-best defense by +12.2% adversarial accuracy against the C&amp;W attack. Our findings highlight the underexplored potential of incorporating external, attacker-unknown context as a robust defense strategy. RobustQuote offers a promising direction for addressing evolving adversarial threats in modern vision models, directly aligning with the critical challenges in adversarial machine learning and cybersecurity.

Distinguishing between alcohol-associated hepatitis (AH) and alcohol-associated cirrhosis (AC) remains a diagnostic challenge. In this study, we used machine learning with transcriptomics and proteomics data from liver tissue and peripheral mononuclear blood cells (PBMCs) to classify patients with alcohol-associated liver disease. The conditions in the study were AH, AC, and healthy controls. We processed 98 PBMC RNAseq samples, 55 PBMC proteomic samples, 48 liver RNAseq samples, and 53 liver proteomic samples. First, we built separate classification and feature selection pipelines for transcriptomics and proteomics data. The liver tissue models were validated in independent liver tissue datasets. Next, we built integrated gene and protein expression models that allowed us to identify combined gene-protein biomarker panels. For liver tissue, we attained 90% nested-cross validation accuracy in our dataset and 82% accuracy in the independent validation dataset using transcriptomic data. We attained 100% nested-cross validation accuracy in our dataset and 61% accuracy in the independent validation dataset using proteomic data. For PBMCs, we attained 83% and 89% accuracy with transcriptomic and proteomic data, respectively. The integration of the two data types resulted in improved classification accuracy for PBMCs, but not liver tissue. We also identified the following gene-protein matches within the gene-protein biomarker panels: CLEC4M-CLC4M, GSTA1-GSTA2 for liver tissue and SELENBP1-SBP1 for PBMCs. In this study, machine learning models had high classification accuracy for both transcriptomics and proteomics data, across liver tissue and PBMCs. The integration of transcriptomics and proteomics into a multi-omics model yielded improvement in classification accuracy for the PBMC data. The set of integrated gene-protein biomarkers for PBMCs show promise toward developing a liquid biopsy for alcohol-associated liver disease.

Accurate and up-to-date land data is essential for effective planning, development, and decision/policy making by the Ministry of Agrarian Affairs and Spatial Planning/National Land Agency. However, land data does not yet have full data accuracy and reliability, thus weakening the usefulness of the land data itself. This study aims to develop a procedural model to improve the quality of land data. This study began by conducting a data classification method, namely conducting a comprehensive inventory of data in the database and archives of land book documents along with measurement letters. Based on the inventory results, a data adjustment method was then done by aligning data between the database and the archives of land book documents and measurement letters. Finally, a method of improving the quality of land data (data quality improvement) was carried out by improving land data to be better. As a form of supervisory control/monitoring, data verification was carried out on data that had been improved using purposive sampling. The results of this study indicate that the systematic application of this method can produce land data with good data accuracy and reliability. Improving the quality of land data results in complete and accurate mapping, and higher data accuracy, although in one village statistically it has not shown positive data, but with the improvement in the quality of land data, it has increased the accuracy of high and detailed data, as well as increased transparency and accountability to the community.

Methane diffusion models are numerous, but their application, pros, and cons need to be summarized urgently, the link between particle size and methane desorption-diffusion is uncertain. Therefore, this paper carries out the methane desorption-diffusion experiments under multi-scale pressure and particle size conditions to study the effect of coal sample particle size on methane desorption-diffusion. Based on MATLAB, different diffusion models were fitted and comparatively analyzed, and the models were improved in view of their deficiencies. The results showed that the cumulative methane desorption amount and rate are inversely related to the particle size of coal. Comparative analysis of the models revealed that diffusion model under the third type of boundary conditions has a low fitting accuracy for full process of methane desorption, whereas it has a high fitting accuracy of 0.9831 for the first 1000 s of desorption data. Unipore diffusion model, bidisperse diffusion model and new model of dynamic diffusion coefficient have high fitting accuracy for the desorption data of high homogeneous coal, reaching above 0.97, but unipore diffusion model has only 0.5434 fitting accuracy for non-homogeneous coal samples. The fitting accuracy of improved classical unipore diffusion model for both homogeneous and non-homogeneous coals reaches 0.99, which is more extensive than that of unipore diffusion model, only take the first term can satisfy the accuracy requirement, so it is more convenient in the practical engineering application, and has certain reference significance for engineering calculations of gas content and prediction of gas protrusion.

Explainability and transparency have emerged as foundational pillars in the secure deployment of artificial intelligence (AI) systems, especially large language models (LLMs). This chapter examines the evolving landscape of explainable AI (XAI) architectures through the lens of cybersecurity, adversarial robustness, and regulatory compliance. The authors survey core XAI methodologies—including LIME, SHAP, mechanistic interpretability, attention attribution, and causal tracing—evaluating their effectiveness against adversarial threats such as jailbreaking, prompt injection, data poisoning, and hallucination exploitation. The dual nature of XAI is critically examined: while transparency mechanisms bolster defense and trust, they simultaneously introduce novel attack surfaces that adversaries can exploit to subvert explanation systems.

Accurate high-resolution atmospheric temperature profiles are essential for precisely characterizing the evolution of the atmosphere and developing numerical forecasts. Atmospheric datasets, such as ERA5 (the fifth-generation ECMWF Reanalysis) and MERRA-2 (the Modern-Era Retrospective Analysis for Research and Applications, Version 2), provide global and continuous temperature profiles, with fine vertical distribution and horizontal resolution. RAOB (Radiosonde Observation) sounding data have high confidence and representativeness and are usually used for data accuracy verification. Due to the difficulty of updating existing products, and the scarcity of research on mesospheric temperature profiles, this work maximizes the high observation accuracy of RAOB data, combines the benefits of ERA5’s horizontal resolution and MERRA-2’s vertical distribution, and employs the optimal interpolation method to combine the data, in order to produce a fused result with high spatial resolution. After converting all of the data to the same spatial distribution, the optimal interpolation method was used to combine the two datasets from separate places and different pressure layers in order to produce the fused results, which had a vertical distribution of 45 layers and a spatial resolution of 0.25°. The fused data’s RMSE and MAE were 6.0 K and 5.0 K lower than those of the MERRA-2 temperature profile data, respectively, and 0.3 K and 0.4 K lower than those of the ERA5 temperature profile data, respectively. The validation, using data from 2019, showed that the fused data exhibits better correlation and data accuracy than the other two datasets, which demonstrated that the fused algorithm can potentially be used to generate reliable datasets for future meteorological research.

Digitalization of enterprise management using Russian software is the main trend in the development of the country's mining sector. Sharp fluctuations in demand and prices in the mineral markets, as well as the development of technical tools for mechanization and automation of technological processes along with a systematic increase in their cost affect the financial performance of mining companies. This determines the need for a rapid adoption of technical, technological and organizational decisions, which, along with the requirements of the company's owners and regulators to obtain prompt, reliable and the most complete information about the production process explains the urgent need for the prompt introduction of digital control systems for mining operations. In the article the authors describe an algorithm to design a digital model of the mining and geological system, including a digital geological model of the deposit, digital models of mine workings and the surface topography, based on the experience of creating three-dimensional digital models of complex-structured coal deposits. The basic software suite for designing a digital management system for mining operations is the MINEFRAME mining and geological information system developed at the Mining Institute of the Kola Scientific Center of the Russian Academy of Sciences. MINEFRAME allows to perform a full range of tasks aimed at creation of a digital 3D model of the mining and geological system for a coal deposit with the possibility to account for the complex bedding in the modelling. High accuracy and correctness of the input geological sampling data is achieved by means of tools to check for errors in the tables of borehole collar coordinates, inclinometry and sampling. The wireframe modelling methods implemented in MINEFRAME provide the possibility to create wireframe models of seams with high detail in the fracture zones.