Knowledge distillation has become a standard technique for compressing large language models into efficient student models,but existing methods often struggle to balance prediction accuracy with explanation quality.Re...Knowledge distillation has become a standard technique for compressing large language models into efficient student models,but existing methods often struggle to balance prediction accuracy with explanation quality.Recent approaches such as Distilling Step-by-Step(DSbS)introduce explanation supervision,yet they apply it in a uniform manner that may not fully exploit the different learning dynamics of prediction and explanation.In this work,we propose a task-structured curriculum learning(TSCL)framework that structures training into three sequential phases:(i)prediction-only,to establish stable feature representations;(ii)joint prediction-explanation,to align task outputs with rationale generation;and(iii)explanation-only,to refine the quality of rationales.This design provides a simple but effective modification to DSbS,requiring no architectural changes and adding negligible training cost.We justify the phase scheduling with ablation studies and convergence analysis,showing that an initial prediction-heavy stage followed by a balanced joint phase improves both stability and explanation alignment.Extensive experiments on five datasets(e-SNLI,ANLI,CommonsenseQA,SVAMP,and MedNLI)demonstrate that TSCL consistently outperforms strong baselines,achieving gains of+1.7-2.6 points in accuracy and 0.8-1.2 in ROUGE-L,corresponding to relative error reductions of up to 21%.Beyond lexical metrics,human evaluation and ERASERstyle faithfulness diagnostics confirm that TSCL produces more faithful and informative explanations.Comparative training curves further reveal faster convergence and lower variance across seeds.Efficiency analysis shows less than 3%overhead in wall-clock training time and no additional inference cost,making the approach practical for realworld deployment.This study demonstrates that a simple task-structured curriculum can significantly improve the effectiveness of knowledge distillation.By separating and sequencing objectives,TSCL achieves a better balance between accuracy,stability,and explanation quality.The framework generalizes across domains,including medical NLI,and offers a principled recipe for future applications in multimodal reasoning and reinforcement learning.展开更多
Breast cancer screening programs rely heavily on mammography for early detection;however,diagnostic performance is strongly affected by inter-reader variability,breast density,and the limitations of conven-tional comp...Breast cancer screening programs rely heavily on mammography for early detection;however,diagnostic performance is strongly affected by inter-reader variability,breast density,and the limitations of conven-tional computer-aided detection systems.Recent advances in deep learning have enabled more robust and scalable solutions for large-scale screening,yet a systematic comparison of modern object detection architectures on nationally representative datasets remains limited.This study presents a comprehensive quantitative comparison of prominent deep learning–based object detection architectures for Artificial Intelligence-assisted mammography analysis using the MammosighTR dataset,developed within the Turkish National Breast Cancer Screening Program.The dataset comprises 12,740 patient cases collected between 2016 and 2022,annotated with BI-RADS categories,breast density levels,and lesion localization labels.A total of 31 models were evaluated,including One-Stage,Two-Stage,and Transformer-based architectures,under a unified experimental framework at both patient and breast levels.The results demonstrate that Two-Stage architectures consistently outperform One-Stage models,achieving approximately 2%–4%higher Macro F1-Scores and more balanced precision–recall trade-offs,with Double-Head R-CNN and Dynamic R-CNN yielding the highest overall performance(Macro F1≈0.84–0.86).This advantage is primarily attributed to the region proposal mechanism and improved class balance inherent to Two-Stage designs.One-Stage detectors exhibited higher sensitivity and faster inference,reaching Recall values above 0.88,but experienced minor reductions in Precision and overall accuracy(≈1%–2%)compared with Two-Stage models.Among Transformer-based architectures,Deformable DEtection TRansformer demonstrated strong robustness and consistency across datasets,achieving Macro F1-Scores comparable to CNN-based detectors(≈0.83–0.85)while exhibiting minimal performance degradation under distributional shifts.Breast density–based analysis revealed increased misclassification rates in medium-density categories(types B and C),whereas Transformer-based architectures maintained more stable performance in high-density type D tissue.These findings quantitatively confirm that both architectural design and tissue characteristics play a decisive role in diagnostic accuracy.Overall,the study provides a reproducible benchmark and highlights the potential of hybrid approaches that combine the accuracy of Two-Stage detectors with the contextual modeling capability of Transformer architectures for clinically reliable breast cancer screening systems.展开更多
Missing data presents a crucial challenge in data analysis,especially in high-dimensional datasets,where missing data often leads to biased conclusions and degraded model performance.In this study,we present a novel a...Missing data presents a crucial challenge in data analysis,especially in high-dimensional datasets,where missing data often leads to biased conclusions and degraded model performance.In this study,we present a novel autoencoder-based imputation framework that integrates a composite loss function to enhance robustness and precision.The proposed loss combines(i)a guided,masked mean squared error focusing on missing entries;(ii)a noise-aware regularization term to improve resilience against data corruption;and(iii)a variance penalty to encourage expressive yet stable reconstructions.We evaluate the proposed model across four missingness mechanisms,such as Missing Completely at Random,Missing at Random,Missing Not at Random,and Missing Not at Random with quantile censorship,under systematically varied feature counts,sample sizes,and missingness ratios ranging from 5%to 60%.Four publicly available real-world datasets(Stroke Prediction,Pima Indians Diabetes,Cardiovascular Disease,and Framingham Heart Study)were used,and the obtained results show that our proposed model consistently outperforms baseline methods,including traditional and deep learning-based techniques.An ablation study reveals the additive value of each component in the loss function.Additionally,we assessed the downstream utility of imputed data through classification tasks,where datasets imputed by the proposed method yielded the highest receiver operating characteristic area under the curve scores across all scenarios.The model demonstrates strong scalability and robustness,improving performance with larger datasets and higher feature counts.These results underscore the capacity of the proposed method to produce not only numerically accurate but also semantically useful imputations,making it a promising solution for robust data recovery in clinical applications.展开更多
Sudden wildfires cause significant global ecological damage.While satellite imagery has advanced early fire detection and mitigation,image-based systems face limitations including high false alarm rates,visual obstruc...Sudden wildfires cause significant global ecological damage.While satellite imagery has advanced early fire detection and mitigation,image-based systems face limitations including high false alarm rates,visual obstructions,and substantial computational demands,especially in complex forest terrains.To address these challenges,this study proposes a novel forest fire detection model utilizing audio classification and machine learning.We developed an audio-based pipeline using real-world environmental sound recordings.Sounds were converted into Mel-spectrograms and classified via a Convolutional Neural Network(CNN),enabling the capture of distinctive fire acoustic signatures(e.g.,crackling,roaring)that are minimally impacted by visual or weather conditions.Internet of Things(IoT)sound sensors were crucial for generating complex environmental parameters to optimize feature extraction.The CNN model achieved high performance in stratified 5-fold cross-validation(92.4%±1.6 accuracy,91.2%±1.8 F1-score)and on test data(94.93%accuracy,93.04%F1-score),with 98.44%precision and 88.32%recall,demonstrating reliability across environmental conditions.These results indicate that the audio-based approach not only improves detection reliability but also markedly reduces computational overhead compared to traditional image-based methods.The findings suggest that acoustic sensing integrated with machine learning offers a powerful,low-cost,and efficient solution for real-time forest fire monitoring in complex,dynamic environments.展开更多
This paper presents a knowledge service system for the domain of agriculture. Three key issues for providing knowledge services are how to improve the access of unstructured and scattered information for the non-speci...This paper presents a knowledge service system for the domain of agriculture. Three key issues for providing knowledge services are how to improve the access of unstructured and scattered information for the non-specialist users, how to provide adequate information to knowledge workers and how to provide the information requiring highly focused and related information. Cyber-Brain has been designed as a platform that combines approaches based on knowledge engineering and language engineering to gather knowledge from various sources and to provide the effective knowledge service. Based on specially designed ontology for practical service scenarios, it can aggregate knowledge from Internet, digital archives, expert, and other resources for providing one-stop-shop knowledge services. The domain specific and task oriented ontology also enables advanced search and allows the system ensures that knowledge service could improve the user benefit. Users are presented with the necessary information closely related to their information need and thus of potential high interest. This paper presents several service scenarios for different end-users and reviews ontology engineering and its life cycle for supporting AOS （Agricultural Ontology Services） Vocbench which is the heart of knowledge services in agriculture domain.展开更多
In this paper, the sticker based DNA computing was used for solving the independent set problem. At first, solution space was constructed by using appropriate DNA memory complexes. We defined a new operation called “...In this paper, the sticker based DNA computing was used for solving the independent set problem. At first, solution space was constructed by using appropriate DNA memory complexes. We defined a new operation called “divide” and applied it in construction of solution space. Then, by application of a sticker based parallel algorithm using biological operations, independent set problem was resolved in polynomial time.展开更多
The ability to accurately predict urban traffic flows is crucial for optimising city operations.Consequently,various methods for forecasting urban traffic have been developed,focusing on analysing historical data to u...The ability to accurately predict urban traffic flows is crucial for optimising city operations.Consequently,various methods for forecasting urban traffic have been developed,focusing on analysing historical data to understand complex mobility patterns.Deep learning techniques,such as graph neural networks(GNNs),are popular for their ability to capture spatio-temporal dependencies.However,these models often become overly complex due to the large number of hyper-parameters involved.In this study,we introduce Dynamic Multi-Graph Spatial-Temporal Graph Neural Ordinary Differential Equation Networks(DMST-GNODE),a framework based on ordinary differential equations(ODEs)that autonomously discovers effective spatial-temporal graph neural network(STGNN)architectures for traffic prediction tasks.The comparative analysis of DMST-GNODE and baseline models indicates that DMST-GNODE model demonstrates superior performance across multiple datasets,consistently achieving the lowest Root Mean Square Error(RMSE)and Mean Absolute Error(MAE)values,alongside the highest accuracy.On the BKK(Bangkok)dataset,it outperformed other models with an RMSE of 3.3165 and an accuracy of 0.9367 for a 20-min interval,maintaining this trend across 40 and 60 min.Similarly,on the PeMS08 dataset,DMST-GNODE achieved the best performance with an RMSE of 19.4863 and an accuracy of 0.9377 at 20 min,demonstrating its effectiveness over longer periods.The Los_Loop dataset results further emphasise this model’s advantage,with an RMSE of 3.3422 and an accuracy of 0.7643 at 20 min,consistently maintaining superiority across all time intervals.These numerical highlights indicate that DMST-GNODE not only outperforms baseline models but also achieves higher accuracy and lower errors across different time intervals and datasets.展开更多
Large Language Models(LLMs)have significantly advanced human-computer interaction by improving natural language understanding and generation.However,their vulnerability to adversarial prompts–carefully designed input...Large Language Models(LLMs)have significantly advanced human-computer interaction by improving natural language understanding and generation.However,their vulnerability to adversarial prompts–carefully designed inputs that manipulate model outputs–presents substantial challenges.This paper introduces a classification-based approach to detect adversarial prompts by utilizing both prompt features and prompt response features.Elevenmachine learning models were evaluated based on key metrics such as accuracy,precision,recall,and F1-score.The results show that the Convolutional Neural Network–Long Short-Term Memory(CNN-LSTM)cascade model delivers the best performance,especially when using prompt features,achieving an accuracy of over 97%in all adversarial scenarios.Furthermore,the Support Vector Machine(SVM)model performed best with prompt response features,particularly excelling in prompt type classification tasks.Classification results revealed that certain types of adversarial attacks,such as“Word Level”and“Adversarial Prefix”,were particularly difficult to detect,as indicated by their low recall and F1-scores.These findings suggest that more subtle manipulations can evade detection mechanisms.In contrast,attacks like“Sentence Level”and“Adversarial Insertion”were easier to identify,due to the model’s effectiveness in recognizing inserted content.Natural Language Processing(NLP)techniques played a critical role by enabling the extraction of semantic and syntactic features from both prompts and their corresponding responses.These insights highlight the importance of combining traditional and deep learning approaches,along with advanced NLP techniques,to build more reliable adversarial prompt detection systems for LLMs.展开更多
Cellulose-based materials have attracted growing interest in the development of advanced energy storage systems owing to their intrinsic sustainability,tunable physicochemical properties,and structural versatility.Thi...Cellulose-based materials have attracted growing interest in the development of advanced energy storage systems owing to their intrinsic sustainability,tunable physicochemical properties,and structural versatility.This review systematically summarizes the key features of cellulose from the perspectives of synthesis,physicochemical characteristics,and structural design,highlighting its unique functionality and adaptability.Furthermore,the roles of cellulose in four critical battery components,i.e.,electrode,solid electrolyte interphase,separator,and electrolyte,are comprehensively discussed,emphasizing the properties aligning with the specific requirements of each component.Finally,potential research directions are proposed to guide future development.This review provides a comprehensive framework for understanding the transformative potential of cellulose in sustainable electrochemical energy storage systems as well as a guideline for future studies.展开更多
The Internet has been enhanced recently by blockchain and Internet of Things(IoT)networks.The Internet of Things is a network of various sensor-equipped devices.It gradually integrates the Internet,sensors,and cloud c...The Internet has been enhanced recently by blockchain and Internet of Things(IoT)networks.The Internet of Things is a network of various sensor-equipped devices.It gradually integrates the Internet,sensors,and cloud computing.Blockchain is based on encryption algorithms,which are shared database technologies on the Internet.Blockchain technology has grown significantly because of its features,such as flexibility,support for integration,anonymity,decentralization,and independent control.Computational nodes in the blockchain network are used to verify online transactions.However,this integration creates scalability,interoperability,and security challenges.Over the last decade,several advancements in blockchain technology have drawn attention fromresearch communities and industries.Blockchain technology helps IoT networks become more reliable and enhance security and privacy.It also removes single points of failure and lowers the cost.In recent years,there has been an increasing amount of literature on IoT and blockchain technology applications.This paper extensively examines the current state of blockchain technologies,focusing specifically on their integration into the Internet of Things.Additionally,it highlights the benefits,drawbacks,and opportunities of recent studies on security issues based on blockchain solutions into categories.The survey examined various research papers fromdifferent types of publications.Also,a review of the other IoT applications has been included,focusing on the security requirements and challenges in IoT-based systems.Future research directions are gathered for the effective integration of Blockchain and IoT.展开更多
Changes in the Atmospheric Electric Field Signal(AEFS)are highly correlated with weather changes,especially with thunderstorm activities.However,little attention has been paid to the ambiguous weather information impl...Changes in the Atmospheric Electric Field Signal(AEFS)are highly correlated with weather changes,especially with thunderstorm activities.However,little attention has been paid to the ambiguous weather information implicit in AEFS changes.In this paper,a Fuzzy C-Means(FCM)clustering method is used for the first time to develop an innovative approach to characterize the weather attributes carried by AEFS.First,a time series dataset is created in the time domain using AEFS attributes.The AEFS-based weather is evaluated according to the time-series Membership Degree(MD)changes obtained by inputting this dataset into the FCM.Second,thunderstorm intensities are reflected by the change in distance from a thunderstorm cloud point charge to an AEF apparatus.Thus,a matching relationship is established between the normalized distance and the thunderstorm dominant MD in the space domain.Finally,the rationality and reliability of the proposed method are verified by combining radar charts and expert experience.The results confirm that this method accurately characterizes the weather attributes and changes in the AEFS,and a negative distance-MD correlation is obtained for the first time.The detection of thunderstorm activity by AEF from the perspective of fuzzy set technology provides a meaningful guidance for interpretable thunderstorms.展开更多
Thunderstorm detection based on the Atmospheric Electric Field(AEF)has evolved from time-domain models to space-domain models.It is especially important to evaluate and determine the particularly Weather Attribute(WA)...Thunderstorm detection based on the Atmospheric Electric Field(AEF)has evolved from time-domain models to space-domain models.It is especially important to evaluate and determine the particularly Weather Attribute(WA),which is directly related to the detection reliability and authenticity.In this paper,a strategy is proposed to integrate three currently competitive WA's evaluation methods.First,a conventional evaluation method based on AEF statistical indicators is selected.Subsequent evaluation approaches include competing AEF-based predicted value intervals,and AEF classification based on fuzzy c-means.Different AEF attributes contribute to a more accurate AEF classification to different degrees.The resulting dynamic weighting applied to these attributes improves the classification accuracy.Each evaluation method is applied to evaluate the WA of a particular AEF,to obtain the corresponding evaluation score.The integration in the proposed strategy takes the form of a score accumulation.Different cumulative score levels correspond to different final WA results.Thunderstorm imaging is performed to visualize thunderstorm activities using those AEFs already evaluated to exhibit thunderstorm attributes.Empirical results confirm that the proposed strategy effectively and reliably images thunderstorms,with a 100%accuracy of WA evaluation.This is the first study to design an integrated thunderstorm detection strategy from a new perspective of WA evaluation,which provides promising solutions for a more reliable and flexible thunderstorm detection.展开更多
Deep neural networks provide accurate results for most applications.However,they need a big dataset to train properly.Providing a big dataset is a significant challenge in most applications.Image augmentation refers t...Deep neural networks provide accurate results for most applications.However,they need a big dataset to train properly.Providing a big dataset is a significant challenge in most applications.Image augmentation refers to techniques that increase the amount of image data.Common operations for image augmentation include changes in illumination,rotation,contrast,size,viewing angle,and others.Recently,Generative Adversarial Networks(GANs)have been employed for image generation.However,like image augmentation methods,GAN approaches can only generate images that are similar to the original images.Therefore,they also cannot generate new classes of data.Texture images presentmore challenges than general images,and generating textures is more complex than creating other types of images.This study proposes a gradient-based deep neural network method that generates a new class of texture.It is possible to rapidly generate new classes of textures using different kernels from pre-trained deep networks.After generating new textures for each class,the number of textures increases through image augmentation.During this process,several techniques are proposed to automatically remove incomplete and similar textures that are created.The proposed method is faster than some well-known generative networks by around 4 to 10 times.In addition,the quality of the generated textures surpasses that of these networks.The proposed method can generate textures that surpass those of someGANs and parametric models in certain image qualitymetrics.It can provide a big texture dataset to train deep networks.A new big texture dataset is created artificially using the proposed method.This dataset is approximately 2 GB in size and comprises 30,000 textures,each 150×150 pixels in size,organized into 600 classes.It is uploaded to the Kaggle site and Google Drive.This dataset is called BigTex.Compared to other texture datasets,the proposed dataset is the largest and can serve as a comprehensive texture dataset for training more powerful deep neural networks and mitigating overfitting.展开更多
Currently, there is a growing belief that putting an IEEE 802.11-like radio into road vehicles can help the drivers to travel more safely. Message dissemination protocols are primordial for safety vehicular applicatio...Currently, there is a growing belief that putting an IEEE 802.11-like radio into road vehicles can help the drivers to travel more safely. Message dissemination protocols are primordial for safety vehicular applications. There are two types of safety messages which may be exchanged between vehicles: alarm and beacon. In this paper we investigate the feasibility of deploying safety applications based on beacon message dissemination through extensive simulation study and pay special attention to the safety requirements. Vehicles are supposed to issue these messages periodically to announce to other vehicles their current situation and use received messages for preventing possible unsafe situations. We evaluate the performance of a single-hop dissemination protocol while taking into account the quality of service (QoS) metrics like delivery rate and delay. We realize that reliability is the main concern in beacon message dissemination. Thus, a new metric named effective range is defined which gives us more accurate facility for evaluating QoS in safety applications specifically. Then, in order to improve the performance, the effects of three parameters including vehicle's transmission range, message transmission's interval time and message payload size are studied. Due to special characteristics of the safety applications, we model the relationship between communication-level QoS and application-level QoS and evaluate them for different classes of safety applications. As a conclusion, the current technology of IEEE 802.11 MAC layer has still some challenges for automatic safety applications but it can provide acceptable QoS to driver assistance safety applications.展开更多
Although real-world experiences show that preparing one image per person is more convenient, most of the appearance-based face recognition methods degrade or fail to work if there is only a single sample per person(SS...Although real-world experiences show that preparing one image per person is more convenient, most of the appearance-based face recognition methods degrade or fail to work if there is only a single sample per person(SSPP). In this work, we introduce a novel supervised learning method called supervised locality preserving multimanifold(SLPMM) for face recognition with SSPP. In SLPMM, two graphs: within-manifold graph and between-manifold graph are made to represent the information inside every manifold and the information among different manifolds, respectively. SLPMM simultaneously maximizes the between-manifold scatter and minimizes the within-manifold scatter which leads to discriminant space by adopting locality preserving projection(LPP) concept. Experimental results on two widely used face databases FERET and AR face database are presented to prove the efficacy of the proposed approach.展开更多
An increasing number of enterprises have adopted cloud computing to manage their important business applications in distributed green cloud(DGC)systems for low response time and high cost-effectiveness in recent years...An increasing number of enterprises have adopted cloud computing to manage their important business applications in distributed green cloud(DGC)systems for low response time and high cost-effectiveness in recent years.Task scheduling and resource allocation in DGCs have gained more attention in both academia and industry as they are costly to manage because of high energy consumption.Many factors in DGCs,e.g.,prices of power grid,and the amount of green energy express strong spatial variations.The dramatic increase of arriving tasks brings a big challenge to minimize the energy cost of a DGC provider in a market where above factors all possess spatial variations.This work adopts a G/G/1 queuing system to analyze the performance of servers in DGCs.Based on it,a single-objective constrained optimization problem is formulated and solved by a proposed simulated-annealing-based bees algorithm(SBA)to find SBA can minimize the energy cost of a DGC provider by optimally allocating tasks of heterogeneous applications among multiple DGCs,and specifying the running speed of each server and the number of powered-on servers in each GC while strictly meeting response time limits of tasks of all applications.Realistic databased experimental results prove that SBA achieves lower energy cost than several benchmark scheduling methods do.展开更多
Hyperspectral images in remote sensing include hundreds of spectral bands that provide valuable information for accurately identify objects.In this paper,a new method of classifying hyperspectral images using spectral...Hyperspectral images in remote sensing include hundreds of spectral bands that provide valuable information for accurately identify objects.In this paper,a new method of classifying hyperspectral images using spectral spatial information has been presented.Here,using the hyperspectral signal subspace identification(HYSIME)method which estimates the signal and noise correlation matrix and selects a subset of eigenvalues for the best representation of the signal subspace in order to minimize the mean square error,subsets from the main sample space have been extracted.After subspace extraction with the help of the HYSIME method,the edge-preserving filtering(EPF),and classification of the hyperspectral subspace using a support vector machine(SVM),results were then merged into the decision-making level using majority rule to create the spectral-spatial classifier.The simulation results showed that the spectral-spatial classifier presented leads to significant improvement in the accuracy and validity of the classification of Indiana,Pavia and Salinas hyperspectral images,such that it can classify these images with 98.79%,98.88% and 97.31% accuracy,respectively.展开更多
A new recommendation method was presented based on memetic algorithm-based clustering. The proposed method was tested on four highly sparse real-world datasets. Its recommendation performance is evaluated and compared...A new recommendation method was presented based on memetic algorithm-based clustering. The proposed method was tested on four highly sparse real-world datasets. Its recommendation performance is evaluated and compared with that of the frequency-based, user-based, item-based, k-means clustering-based, and genetic algorithm-based methods in terms of precision, recall, and F1 score. The results show that the proposed method yields better performance under the new user cold-start problem when each of new active users selects only one or two items into the basket. The average F1 scores on all four datasets are improved by 225.0%, 61.6%, 54.6%, 49.3%, 28.8%, and 6.3% over the frequency-based, user-based, item-based, k-means clustering-based, and two genetic algorithm-based methods, respectively.展开更多
In convolutional neural networks,pooling methods are used to reduce both the size of the data and the number of parameters after the convolution of the models.These methods reduce the computational amount of convoluti...In convolutional neural networks,pooling methods are used to reduce both the size of the data and the number of parameters after the convolution of the models.These methods reduce the computational amount of convolutional neural networks,making the neural network more efficient.Maximum pooling,average pooling,and minimum pooling methods are generally used in convolutional neural networks.However,these pooling methods are not suitable for all datasets used in neural network applications.In this study,a new pooling approach to the literature is proposed to increase the efficiency and success rates of convolutional neural networks.This method,which we call MAM(Maximum Average Minimum)pooling,is more interactive than other traditional maximum pooling,average pooling,and minimum pooling methods and reduces data loss by calculating the more appropriate pixel value.The proposed MAM pooling method increases the performance of the neural network by calculating the optimal value during the training of convolutional neural networks.To determine the success accuracy of the proposed MAM pooling method and compare it with other traditional pooling methods,training was carried out on the LeNet-5 model using CIFAR-10,CIFAR-100,and MNIST datasets.According to the results obtained,the proposed MAM pooling method performed better than the maximum pooling,average pooling,and minimum pooling methods in all pool sizes on three different datasets.展开更多
文摘Knowledge distillation has become a standard technique for compressing large language models into efficient student models,but existing methods often struggle to balance prediction accuracy with explanation quality.Recent approaches such as Distilling Step-by-Step(DSbS)introduce explanation supervision,yet they apply it in a uniform manner that may not fully exploit the different learning dynamics of prediction and explanation.In this work,we propose a task-structured curriculum learning(TSCL)framework that structures training into three sequential phases:(i)prediction-only,to establish stable feature representations;(ii)joint prediction-explanation,to align task outputs with rationale generation;and(iii)explanation-only,to refine the quality of rationales.This design provides a simple but effective modification to DSbS,requiring no architectural changes and adding negligible training cost.We justify the phase scheduling with ablation studies and convergence analysis,showing that an initial prediction-heavy stage followed by a balanced joint phase improves both stability and explanation alignment.Extensive experiments on five datasets(e-SNLI,ANLI,CommonsenseQA,SVAMP,and MedNLI)demonstrate that TSCL consistently outperforms strong baselines,achieving gains of+1.7-2.6 points in accuracy and 0.8-1.2 in ROUGE-L,corresponding to relative error reductions of up to 21%.Beyond lexical metrics,human evaluation and ERASERstyle faithfulness diagnostics confirm that TSCL produces more faithful and informative explanations.Comparative training curves further reveal faster convergence and lower variance across seeds.Efficiency analysis shows less than 3%overhead in wall-clock training time and no additional inference cost,making the approach practical for realworld deployment.This study demonstrates that a simple task-structured curriculum can significantly improve the effectiveness of knowledge distillation.By separating and sequencing objectives,TSCL achieves a better balance between accuracy,stability,and explanation quality.The framework generalizes across domains,including medical NLI,and offers a principled recipe for future applications in multimodal reasoning and reinforcement learning.
文摘Breast cancer screening programs rely heavily on mammography for early detection;however,diagnostic performance is strongly affected by inter-reader variability,breast density,and the limitations of conven-tional computer-aided detection systems.Recent advances in deep learning have enabled more robust and scalable solutions for large-scale screening,yet a systematic comparison of modern object detection architectures on nationally representative datasets remains limited.This study presents a comprehensive quantitative comparison of prominent deep learning–based object detection architectures for Artificial Intelligence-assisted mammography analysis using the MammosighTR dataset,developed within the Turkish National Breast Cancer Screening Program.The dataset comprises 12,740 patient cases collected between 2016 and 2022,annotated with BI-RADS categories,breast density levels,and lesion localization labels.A total of 31 models were evaluated,including One-Stage,Two-Stage,and Transformer-based architectures,under a unified experimental framework at both patient and breast levels.The results demonstrate that Two-Stage architectures consistently outperform One-Stage models,achieving approximately 2%–4%higher Macro F1-Scores and more balanced precision–recall trade-offs,with Double-Head R-CNN and Dynamic R-CNN yielding the highest overall performance(Macro F1≈0.84–0.86).This advantage is primarily attributed to the region proposal mechanism and improved class balance inherent to Two-Stage designs.One-Stage detectors exhibited higher sensitivity and faster inference,reaching Recall values above 0.88,but experienced minor reductions in Precision and overall accuracy(≈1%–2%)compared with Two-Stage models.Among Transformer-based architectures,Deformable DEtection TRansformer demonstrated strong robustness and consistency across datasets,achieving Macro F1-Scores comparable to CNN-based detectors(≈0.83–0.85)while exhibiting minimal performance degradation under distributional shifts.Breast density–based analysis revealed increased misclassification rates in medium-density categories(types B and C),whereas Transformer-based architectures maintained more stable performance in high-density type D tissue.These findings quantitatively confirm that both architectural design and tissue characteristics play a decisive role in diagnostic accuracy.Overall,the study provides a reproducible benchmark and highlights the potential of hybrid approaches that combine the accuracy of Two-Stage detectors with the contextual modeling capability of Transformer architectures for clinically reliable breast cancer screening systems.
文摘Missing data presents a crucial challenge in data analysis,especially in high-dimensional datasets,where missing data often leads to biased conclusions and degraded model performance.In this study,we present a novel autoencoder-based imputation framework that integrates a composite loss function to enhance robustness and precision.The proposed loss combines(i)a guided,masked mean squared error focusing on missing entries;(ii)a noise-aware regularization term to improve resilience against data corruption;and(iii)a variance penalty to encourage expressive yet stable reconstructions.We evaluate the proposed model across four missingness mechanisms,such as Missing Completely at Random,Missing at Random,Missing Not at Random,and Missing Not at Random with quantile censorship,under systematically varied feature counts,sample sizes,and missingness ratios ranging from 5%to 60%.Four publicly available real-world datasets(Stroke Prediction,Pima Indians Diabetes,Cardiovascular Disease,and Framingham Heart Study)were used,and the obtained results show that our proposed model consistently outperforms baseline methods,including traditional and deep learning-based techniques.An ablation study reveals the additive value of each component in the loss function.Additionally,we assessed the downstream utility of imputed data through classification tasks,where datasets imputed by the proposed method yielded the highest receiver operating characteristic area under the curve scores across all scenarios.The model demonstrates strong scalability and robustness,improving performance with larger datasets and higher feature counts.These results underscore the capacity of the proposed method to produce not only numerically accurate but also semantically useful imputations,making it a promising solution for robust data recovery in clinical applications.
基金funded by the Directorate of Research and Community Service,Directorate General of Research and Development,Ministry of Higher Education,Science and Technologyin accordance with the Implementation Contract for the Operational Assistance Program for State Universities,Research Program Number:109/C3/DT.05.00/PL/2025.
文摘Sudden wildfires cause significant global ecological damage.While satellite imagery has advanced early fire detection and mitigation,image-based systems face limitations including high false alarm rates,visual obstructions,and substantial computational demands,especially in complex forest terrains.To address these challenges,this study proposes a novel forest fire detection model utilizing audio classification and machine learning.We developed an audio-based pipeline using real-world environmental sound recordings.Sounds were converted into Mel-spectrograms and classified via a Convolutional Neural Network(CNN),enabling the capture of distinctive fire acoustic signatures(e.g.,crackling,roaring)that are minimally impacted by visual or weather conditions.Internet of Things(IoT)sound sensors were crucial for generating complex environmental parameters to optimize feature extraction.The CNN model achieved high performance in stratified 5-fold cross-validation(92.4%±1.6 accuracy,91.2%±1.8 F1-score)and on test data(94.93%accuracy,93.04%F1-score),with 98.44%precision and 88.32%recall,demonstrating reliability across environmental conditions.These results indicate that the audio-based approach not only improves detection reliability but also markedly reduces computational overhead compared to traditional image-based methods.The findings suggest that acoustic sensing integrated with machine learning offers a powerful,low-cost,and efficient solution for real-time forest fire monitoring in complex,dynamic environments.
文摘This paper presents a knowledge service system for the domain of agriculture. Three key issues for providing knowledge services are how to improve the access of unstructured and scattered information for the non-specialist users, how to provide adequate information to knowledge workers and how to provide the information requiring highly focused and related information. Cyber-Brain has been designed as a platform that combines approaches based on knowledge engineering and language engineering to gather knowledge from various sources and to provide the effective knowledge service. Based on specially designed ontology for practical service scenarios, it can aggregate knowledge from Internet, digital archives, expert, and other resources for providing one-stop-shop knowledge services. The domain specific and task oriented ontology also enables advanced search and allows the system ensures that knowledge service could improve the user benefit. Users are presented with the necessary information closely related to their information need and thus of potential high interest. This paper presents several service scenarios for different end-users and reviews ontology engineering and its life cycle for supporting AOS （Agricultural Ontology Services） Vocbench which is the heart of knowledge services in agriculture domain.
文摘In this paper, the sticker based DNA computing was used for solving the independent set problem. At first, solution space was constructed by using appropriate DNA memory complexes. We defined a new operation called “divide” and applied it in construction of solution space. Then, by application of a sticker based parallel algorithm using biological operations, independent set problem was resolved in polynomial time.
文摘The ability to accurately predict urban traffic flows is crucial for optimising city operations.Consequently,various methods for forecasting urban traffic have been developed,focusing on analysing historical data to understand complex mobility patterns.Deep learning techniques,such as graph neural networks(GNNs),are popular for their ability to capture spatio-temporal dependencies.However,these models often become overly complex due to the large number of hyper-parameters involved.In this study,we introduce Dynamic Multi-Graph Spatial-Temporal Graph Neural Ordinary Differential Equation Networks(DMST-GNODE),a framework based on ordinary differential equations(ODEs)that autonomously discovers effective spatial-temporal graph neural network(STGNN)architectures for traffic prediction tasks.The comparative analysis of DMST-GNODE and baseline models indicates that DMST-GNODE model demonstrates superior performance across multiple datasets,consistently achieving the lowest Root Mean Square Error(RMSE)and Mean Absolute Error(MAE)values,alongside the highest accuracy.On the BKK(Bangkok)dataset,it outperformed other models with an RMSE of 3.3165 and an accuracy of 0.9367 for a 20-min interval,maintaining this trend across 40 and 60 min.Similarly,on the PeMS08 dataset,DMST-GNODE achieved the best performance with an RMSE of 19.4863 and an accuracy of 0.9377 at 20 min,demonstrating its effectiveness over longer periods.The Los_Loop dataset results further emphasise this model’s advantage,with an RMSE of 3.3422 and an accuracy of 0.7643 at 20 min,consistently maintaining superiority across all time intervals.These numerical highlights indicate that DMST-GNODE not only outperforms baseline models but also achieves higher accuracy and lower errors across different time intervals and datasets.
文摘Large Language Models(LLMs)have significantly advanced human-computer interaction by improving natural language understanding and generation.However,their vulnerability to adversarial prompts–carefully designed inputs that manipulate model outputs–presents substantial challenges.This paper introduces a classification-based approach to detect adversarial prompts by utilizing both prompt features and prompt response features.Elevenmachine learning models were evaluated based on key metrics such as accuracy,precision,recall,and F1-score.The results show that the Convolutional Neural Network–Long Short-Term Memory(CNN-LSTM)cascade model delivers the best performance,especially when using prompt features,achieving an accuracy of over 97%in all adversarial scenarios.Furthermore,the Support Vector Machine(SVM)model performed best with prompt response features,particularly excelling in prompt type classification tasks.Classification results revealed that certain types of adversarial attacks,such as“Word Level”and“Adversarial Prefix”,were particularly difficult to detect,as indicated by their low recall and F1-scores.These findings suggest that more subtle manipulations can evade detection mechanisms.In contrast,attacks like“Sentence Level”and“Adversarial Insertion”were easier to identify,due to the model’s effectiveness in recognizing inserted content.Natural Language Processing(NLP)techniques played a critical role by enabling the extraction of semantic and syntactic features from both prompts and their corresponding responses.These insights highlight the importance of combining traditional and deep learning approaches,along with advanced NLP techniques,to build more reliable adversarial prompt detection systems for LLMs.
基金the funding support from University of Macao(Grant No.SRG2024-00034-IAPME)The Science and Technology Development Fund from Macao SAR(FDCT)(Grant No.0002/2024/TFP).
文摘Cellulose-based materials have attracted growing interest in the development of advanced energy storage systems owing to their intrinsic sustainability,tunable physicochemical properties,and structural versatility.This review systematically summarizes the key features of cellulose from the perspectives of synthesis,physicochemical characteristics,and structural design,highlighting its unique functionality and adaptability.Furthermore,the roles of cellulose in four critical battery components,i.e.,electrode,solid electrolyte interphase,separator,and electrolyte,are comprehensively discussed,emphasizing the properties aligning with the specific requirements of each component.Finally,potential research directions are proposed to guide future development.This review provides a comprehensive framework for understanding the transformative potential of cellulose in sustainable electrochemical energy storage systems as well as a guideline for future studies.
文摘The Internet has been enhanced recently by blockchain and Internet of Things(IoT)networks.The Internet of Things is a network of various sensor-equipped devices.It gradually integrates the Internet,sensors,and cloud computing.Blockchain is based on encryption algorithms,which are shared database technologies on the Internet.Blockchain technology has grown significantly because of its features,such as flexibility,support for integration,anonymity,decentralization,and independent control.Computational nodes in the blockchain network are used to verify online transactions.However,this integration creates scalability,interoperability,and security challenges.Over the last decade,several advancements in blockchain technology have drawn attention fromresearch communities and industries.Blockchain technology helps IoT networks become more reliable and enhance security and privacy.It also removes single points of failure and lowers the cost.In recent years,there has been an increasing amount of literature on IoT and blockchain technology applications.This paper extensively examines the current state of blockchain technologies,focusing specifically on their integration into the Internet of Things.Additionally,it highlights the benefits,drawbacks,and opportunities of recent studies on security issues based on blockchain solutions into categories.The survey examined various research papers fromdifferent types of publications.Also,a review of the other IoT applications has been included,focusing on the security requirements and challenges in IoT-based systems.Future research directions are gathered for the effective integration of Blockchain and IoT.
基金supported in part by the National Natural Science Foundation of China under Grant 62171228in part by the National Key R&D Program of China under Grant 2021YFE0105500in part by the Program of China Scholarship Council under Grant 202209040027。
文摘Changes in the Atmospheric Electric Field Signal(AEFS)are highly correlated with weather changes,especially with thunderstorm activities.However,little attention has been paid to the ambiguous weather information implicit in AEFS changes.In this paper,a Fuzzy C-Means(FCM)clustering method is used for the first time to develop an innovative approach to characterize the weather attributes carried by AEFS.First,a time series dataset is created in the time domain using AEFS attributes.The AEFS-based weather is evaluated according to the time-series Membership Degree(MD)changes obtained by inputting this dataset into the FCM.Second,thunderstorm intensities are reflected by the change in distance from a thunderstorm cloud point charge to an AEF apparatus.Thus,a matching relationship is established between the normalized distance and the thunderstorm dominant MD in the space domain.Finally,the rationality and reliability of the proposed method are verified by combining radar charts and expert experience.The results confirm that this method accurately characterizes the weather attributes and changes in the AEFS,and a negative distance-MD correlation is obtained for the first time.The detection of thunderstorm activity by AEF from the perspective of fuzzy set technology provides a meaningful guidance for interpretable thunderstorms.
基金supported in part by the National Natural Science Foundation of China under Grant 62171228in part by the National Key R&D Program of China under Grant 2021YFE0105500in part by the Program of China Scholarship Council under Grant 202209040027。
文摘Thunderstorm detection based on the Atmospheric Electric Field(AEF)has evolved from time-domain models to space-domain models.It is especially important to evaluate and determine the particularly Weather Attribute(WA),which is directly related to the detection reliability and authenticity.In this paper,a strategy is proposed to integrate three currently competitive WA's evaluation methods.First,a conventional evaluation method based on AEF statistical indicators is selected.Subsequent evaluation approaches include competing AEF-based predicted value intervals,and AEF classification based on fuzzy c-means.Different AEF attributes contribute to a more accurate AEF classification to different degrees.The resulting dynamic weighting applied to these attributes improves the classification accuracy.Each evaluation method is applied to evaluate the WA of a particular AEF,to obtain the corresponding evaluation score.The integration in the proposed strategy takes the form of a score accumulation.Different cumulative score levels correspond to different final WA results.Thunderstorm imaging is performed to visualize thunderstorm activities using those AEFs already evaluated to exhibit thunderstorm attributes.Empirical results confirm that the proposed strategy effectively and reliably images thunderstorms,with a 100%accuracy of WA evaluation.This is the first study to design an integrated thunderstorm detection strategy from a new perspective of WA evaluation,which provides promising solutions for a more reliable and flexible thunderstorm detection.
基金supported via funding from Prince Sattam bin Abdulaziz University(PSAU/2025/R/1446)Princess Nourah bint Abdulrahman University(PNURSP2025R300)Prince Sultan University.
文摘Deep neural networks provide accurate results for most applications.However,they need a big dataset to train properly.Providing a big dataset is a significant challenge in most applications.Image augmentation refers to techniques that increase the amount of image data.Common operations for image augmentation include changes in illumination,rotation,contrast,size,viewing angle,and others.Recently,Generative Adversarial Networks(GANs)have been employed for image generation.However,like image augmentation methods,GAN approaches can only generate images that are similar to the original images.Therefore,they also cannot generate new classes of data.Texture images presentmore challenges than general images,and generating textures is more complex than creating other types of images.This study proposes a gradient-based deep neural network method that generates a new class of texture.It is possible to rapidly generate new classes of textures using different kernels from pre-trained deep networks.After generating new textures for each class,the number of textures increases through image augmentation.During this process,several techniques are proposed to automatically remove incomplete and similar textures that are created.The proposed method is faster than some well-known generative networks by around 4 to 10 times.In addition,the quality of the generated textures surpasses that of these networks.The proposed method can generate textures that surpass those of someGANs and parametric models in certain image qualitymetrics.It can provide a big texture dataset to train deep networks.A new big texture dataset is created artificially using the proposed method.This dataset is approximately 2 GB in size and comprises 30,000 textures,each 150×150 pixels in size,organized into 600 classes.It is uploaded to the Kaggle site and Google Drive.This dataset is called BigTex.Compared to other texture datasets,the proposed dataset is the largest and can serve as a comprehensive texture dataset for training more powerful deep neural networks and mitigating overfitting.
基金the Iran Telecommunication Research Center (ITRC)
文摘Currently, there is a growing belief that putting an IEEE 802.11-like radio into road vehicles can help the drivers to travel more safely. Message dissemination protocols are primordial for safety vehicular applications. There are two types of safety messages which may be exchanged between vehicles: alarm and beacon. In this paper we investigate the feasibility of deploying safety applications based on beacon message dissemination through extensive simulation study and pay special attention to the safety requirements. Vehicles are supposed to issue these messages periodically to announce to other vehicles their current situation and use received messages for preventing possible unsafe situations. We evaluate the performance of a single-hop dissemination protocol while taking into account the quality of service (QoS) metrics like delivery rate and delay. We realize that reliability is the main concern in beacon message dissemination. Thus, a new metric named effective range is defined which gives us more accurate facility for evaluating QoS in safety applications specifically. Then, in order to improve the performance, the effects of three parameters including vehicle's transmission range, message transmission's interval time and message payload size are studied. Due to special characteristics of the safety applications, we model the relationship between communication-level QoS and application-level QoS and evaluate them for different classes of safety applications. As a conclusion, the current technology of IEEE 802.11 MAC layer has still some challenges for automatic safety applications but it can provide acceptable QoS to driver assistance safety applications.
文摘Although real-world experiences show that preparing one image per person is more convenient, most of the appearance-based face recognition methods degrade or fail to work if there is only a single sample per person(SSPP). In this work, we introduce a novel supervised learning method called supervised locality preserving multimanifold(SLPMM) for face recognition with SSPP. In SLPMM, two graphs: within-manifold graph and between-manifold graph are made to represent the information inside every manifold and the information among different manifolds, respectively. SLPMM simultaneously maximizes the between-manifold scatter and minimizes the within-manifold scatter which leads to discriminant space by adopting locality preserving projection(LPP) concept. Experimental results on two widely used face databases FERET and AR face database are presented to prove the efficacy of the proposed approach.
基金supported in part by the National Natural Science Foundation of China(61802015,61703011)the Major Science and Technology Program for Water Pollution Control and Treatment of China(2018ZX07111005)+1 种基金the National Defense Pre-Research Foundation of China(41401020401,41401050102)the Deanship of Scientific Research(DSR)at King Abdulaziz University,Jeddah(D-422-135-1441)。
文摘An increasing number of enterprises have adopted cloud computing to manage their important business applications in distributed green cloud(DGC)systems for low response time and high cost-effectiveness in recent years.Task scheduling and resource allocation in DGCs have gained more attention in both academia and industry as they are costly to manage because of high energy consumption.Many factors in DGCs,e.g.,prices of power grid,and the amount of green energy express strong spatial variations.The dramatic increase of arriving tasks brings a big challenge to minimize the energy cost of a DGC provider in a market where above factors all possess spatial variations.This work adopts a G/G/1 queuing system to analyze the performance of servers in DGCs.Based on it,a single-objective constrained optimization problem is formulated and solved by a proposed simulated-annealing-based bees algorithm(SBA)to find SBA can minimize the energy cost of a DGC provider by optimally allocating tasks of heterogeneous applications among multiple DGCs,and specifying the running speed of each server and the number of powered-on servers in each GC while strictly meeting response time limits of tasks of all applications.Realistic databased experimental results prove that SBA achieves lower energy cost than several benchmark scheduling methods do.
文摘Hyperspectral images in remote sensing include hundreds of spectral bands that provide valuable information for accurately identify objects.In this paper,a new method of classifying hyperspectral images using spectral spatial information has been presented.Here,using the hyperspectral signal subspace identification(HYSIME)method which estimates the signal and noise correlation matrix and selects a subset of eigenvalues for the best representation of the signal subspace in order to minimize the mean square error,subsets from the main sample space have been extracted.After subspace extraction with the help of the HYSIME method,the edge-preserving filtering(EPF),and classification of the hyperspectral subspace using a support vector machine(SVM),results were then merged into the decision-making level using majority rule to create the spectral-spatial classifier.The simulation results showed that the spectral-spatial classifier presented leads to significant improvement in the accuracy and validity of the classification of Indiana,Pavia and Salinas hyperspectral images,such that it can classify these images with 98.79%,98.88% and 97.31% accuracy,respectively.
基金supporting by grant fund under the Strategic Scholarships for Frontier Research Network for the PhD Program Thai Doctoral degree
文摘A new recommendation method was presented based on memetic algorithm-based clustering. The proposed method was tested on four highly sparse real-world datasets. Its recommendation performance is evaluated and compared with that of the frequency-based, user-based, item-based, k-means clustering-based, and genetic algorithm-based methods in terms of precision, recall, and F1 score. The results show that the proposed method yields better performance under the new user cold-start problem when each of new active users selects only one or two items into the basket. The average F1 scores on all four datasets are improved by 225.0%, 61.6%, 54.6%, 49.3%, 28.8%, and 6.3% over the frequency-based, user-based, item-based, k-means clustering-based, and two genetic algorithm-based methods, respectively.
文摘In convolutional neural networks,pooling methods are used to reduce both the size of the data and the number of parameters after the convolution of the models.These methods reduce the computational amount of convolutional neural networks,making the neural network more efficient.Maximum pooling,average pooling,and minimum pooling methods are generally used in convolutional neural networks.However,these pooling methods are not suitable for all datasets used in neural network applications.In this study,a new pooling approach to the literature is proposed to increase the efficiency and success rates of convolutional neural networks.This method,which we call MAM(Maximum Average Minimum)pooling,is more interactive than other traditional maximum pooling,average pooling,and minimum pooling methods and reduces data loss by calculating the more appropriate pixel value.The proposed MAM pooling method increases the performance of the neural network by calculating the optimal value during the training of convolutional neural networks.To determine the success accuracy of the proposed MAM pooling method and compare it with other traditional pooling methods,training was carried out on the LeNet-5 model using CIFAR-10,CIFAR-100,and MNIST datasets.According to the results obtained,the proposed MAM pooling method performed better than the maximum pooling,average pooling,and minimum pooling methods in all pool sizes on three different datasets.
