With the prevalence of renewable distributed energy resources(DERs)such as photovoltaics(PVs),modern active distribution networks(ADNs)suffer from voltage deviation and power quality issues.However,traditional voltage...With the prevalence of renewable distributed energy resources(DERs)such as photovoltaics(PVs),modern active distribution networks(ADNs)suffer from voltage deviation and power quality issues.However,traditional voltage control methods often face a trade-off between efficiency and effectiveness,and rarely ensure robust voltage safety under typical state perturbations in practical distribution grids.In this paper,a robust model-free voltage regulation approach is proposed which simultaneously takes security and robustness into account.In this context,the voltage control problem is formulated as a constrained Markov decision process(CMDP).A safety-augmented multiagent deep deterministic policy gradient(MADDPG)algorithm is the trained to enable real-time collaborative optimization of ADNs,aiming to maintain nodal voltages within safe operational limits while minimizing total line losses.Moreover,a robust regulation loss is introduced to ensure reliable performance under various state perturbations in practical voltage controls.The proposed regulation algorithm effectively balance efficiency,safety,and robustness,and also demonstrates potential for generalizing these characteristics to other applications.Numerical studies vali-date the robustness of the proposed method under varying state perturbations on the IEEE test cases and the optimal integrated control performance when compared to other benchmarks.展开更多
The expansion of 5G-enabled Internet of Things(IoT)networks,while enabling transformative applications,significantly increases the attack surface and necessitates security solutions that extend beyond traditional intr...The expansion of 5G-enabled Internet of Things(IoT)networks,while enabling transformative applications,significantly increases the attack surface and necessitates security solutions that extend beyond traditional intrusion detection.Existing intrusion detection systems(IDSs)mainly operate in an open-loop manner,excelling at classification but lacking the ability for autonomous,safety-aware remediation.This gap is particularly critical in 5G environments,where manual intervention is too slow and naive automation can lead to severe service disruptions.To address this issue,we propose a novel Self-Healing Intrusion Detection System(SH-IDS)framework that develops a closed-loop cyber defense mechanism.The main technical contribution is the integration of a deep neural networkbased threat detector,which offers uncertainty-quantified predictions,with a safety-aware reinforcement learning(RL)engine formulated as a Constrained Markov Decision Process(CMDP).The CMDP explicitly models operational safety as cost constraints,and a new runtime safety shield actively adjusts any unsafe action proposed by the RL agent to the nearest safe alternative,ensuring operational integrity.Additionally,we introduce a composite utility function for the comprehensive evaluation of the system.Empirical analysis on the 5G-NIDD dataset demonstrates the superior performance of our framework:the detector achieves 98.26%accuracy,while the safe RL agent learns effective mitigation policies.Importantly,the safety shield blocked up to 70 unsafe actions under strict constraints,and analysis of the learned Q-tables confirms that the agent internalizes safety,avoiding overly disruptive actions,such as isolating nodes for minor threats.The system also maintains high efficiency with a compact model size of 121.7 KB and sub-millisecond latency,confirming its practical deployability for real-time 5G-IoT security.展开更多
The real-time AC optimal power flow(OPF)problem is a key issue in making fast and accurate decisions to ensure the safety and economy of power systems.With the rapid development of renewable energies,the fluctuation h...The real-time AC optimal power flow(OPF)problem is a key issue in making fast and accurate decisions to ensure the safety and economy of power systems.With the rapid development of renewable energies,the fluctuation has grown more vibrant,thus a novel approach called safe deep reinforcement learning is proposed in this paper.Herein,the real-time ACOPF problem is modeled as a constrained Markov decision process,and primal-dual optimization(PDO)based proximal policy optimization(PPO)is used to learn the optimal generator outputs in the primal domain and security constraints in the dual domain,which avoids manually selecting a trade-off between penalties for constraint violations and rewards for the economy.Before training,behavior cloning clones the expert experience into the initial weights of neural networks.Moreover,multiprocessing training is utilized to accelerate the training speed.Case studies are conducted on the IEEE 118-bus system and the modified IEEE 118-bus system.Compared with other methods,the experimental results show that the proposed method can achieve security and near-optimal economic goals by fast calculating the real-time ACOPF problem.展开更多
Grasping is one of the most fundamental operations in modern robotics applications.While deep rein-forcement learning(DRL)has demonstrated strong potential in robotics,there is too much emphasis on maximizing the cumu...Grasping is one of the most fundamental operations in modern robotics applications.While deep rein-forcement learning(DRL)has demonstrated strong potential in robotics,there is too much emphasis on maximizing the cumulative reward in executing tasks,and the potential safety risks are often ignored.In this paper,an optimization method based on safe reinforcement learning(Safe RL)is proposed to address the robotic grasping problem under safety constraints.Specifically,considering the obstacle avoidance constraints of the system,the grasping problem of the manipulator is modeled as a Constrained Markov Decision Process(CMDP).The Lagrange multiplier and a dynamic weighted mechanism are introduced into the Proximal Policy Optimization(PPO)framework,leading to the development of the dynamic weighted Lagrange PPO(DWL-PPO)algorithm.The behavior of violating safety constraints is punished while the policy is optimized in this proposed method.In addition,the orientation control of the end-effector is included in the reward function,and a compound reward function adapted to changes in pose is designed.Ultimately,the efficacy and advantages of the suggested method are proved by extensive training and testing in the Pybullet simulator.The results of grasping experiments reveal that the recommended approach provides superior safety and efficiency compared with other advanced RL methods and achieves a good trade-off between model learning and risk aversion.展开更多
Batteries play a crucial role in the storage and application of sustainable energy,yet their inherent safety risks are non-negligible.Traditional monitoring methods often suffer from high costs,time consumption,and li...Batteries play a crucial role in the storage and application of sustainable energy,yet their inherent safety risks are non-negligible.Traditional monitoring methods often suffer from high costs,time consumption,and limited scalability,making it increasingly difficult to meet the evolving demands of modern society.In this context,recent advancements in machine learning technology have emerged as a promising solution for predicting and monitoring battery states,offering innovative approaches to battery management systems(BMS).By transforming raw operational data into actionable insights,machine learning has shifted the paradigm from reactive to predictive battery safety management,significantly enhancing system reliability and risk mitigation capabilities.This review delves into the implementation of machine learning in battery state prediction,including dataset selection,feature extraction,and model training.It also highlights the latest progress of these models in key applications such as state of health(SOH),state of charge(SOC),thermal runaway warning,fault detection,and remaining useful life(RUL).Finally,we critically examined the challenges and opportunities associated with leveraging machine learning to improve battery safety and performance,providing a comprehensive perspective for future research in this rapidly advancing field.展开更多
Low visibility conditions,particularly those caused by fog,significantly affect road safety and reduce drivers’ability to see ahead clearly.The conventional approaches used to address this problem primarily rely on i...Low visibility conditions,particularly those caused by fog,significantly affect road safety and reduce drivers’ability to see ahead clearly.The conventional approaches used to address this problem primarily rely on instrument-based and fixed-threshold-based theoretical frameworks,which face challenges in adaptability and demonstrate lower performance under varying environmental conditions.To overcome these challenges,we propose a real-time visibility estimation model that leverages roadside CCTV cameras to monitor and identify visibility levels under different weather conditions.The proposedmethod begins by identifying specific regions of interest(ROI)in the CCTVimages and focuses on extracting specific features such as the number of lines and contours detected within these regions.These features are then provided as an input to the proposed hierarchical clusteringmodel,which classifies them into different visibility levels without the need for predefined rules and threshold values.In the proposed approach,we used two different distance similaritymetrics,namely dynamic time warping(DTW)and Euclidean distance,alongside the proposed hierarchical clustering model and noted its performance in terms of numerous evaluation measures.The proposed model achieved an average accuracy of 97.81%,precision of 91.31%,recall of 91.25%,and F1-score of 91.27% using theDTWdistancemetric.We also conducted experiments for other deep learning(DL)-based models used in the literature and compared their performances with the proposed model.The experimental results demonstrate that the proposedmodel ismore adaptable and consistent compared to themethods used in the literature.The proposedmethod provides drivers real-time and accurate visibility information and enhances road safety during low visibility conditions.展开更多
The uplift resistance of the soil overlying shield tunnels significantly impacts their anti-floating stability.However,research on uplift resistance concerning special-shaped shield tunnels is limited.This study combi...The uplift resistance of the soil overlying shield tunnels significantly impacts their anti-floating stability.However,research on uplift resistance concerning special-shaped shield tunnels is limited.This study combines numerical simulation with machine learning techniques to explore this issue.It presents a summary of special-shaped tunnel geometries and introduces a shape coefficient.Through the finite element software,Plaxis3D,the study simulates six key parameters—shape coefficient,burial depth ratio,tunnel’s longest horizontal length,internal friction angle,cohesion,and soil submerged bulk density—that impact uplift resistance across different conditions.Employing XGBoost and ANN methods,the feature importance of each parameter was analyzed based on the numerical simulation results.The findings demonstrate that a tunnel shape more closely resembling a circle leads to reduced uplift resistance in the overlying soil,whereas other parameters exhibit the contrary effects.Furthermore,the study reveals a diminishing trend in the feature importance of buried depth ratio,internal friction angle,tunnel longest horizontal length,cohesion,soil submerged bulk density,and shape coefficient in influencing uplift resistance.展开更多
Countries around the world have been making efforts to reduce pollutant emissions. However, the response of global black carbon(BC) aging to emission changes remains unclear. Using the Community Atmosphere Model versi...Countries around the world have been making efforts to reduce pollutant emissions. However, the response of global black carbon(BC) aging to emission changes remains unclear. Using the Community Atmosphere Model version 6 with a machine-learning-integrated four-mode version of the Modal Aerosol Module, we quantify global BC aging responses to emission reductions for 2011–2018 and for 2050 and 2100 under carbon neutrality. During 2011–18, global trends in BC aging degree(mass ratio of coatings to BC, R_(BC)) exhibited marked regional disparities, with a significant increase in China(5.4% yr^(-1)), which contrasts with minimal changes in the USA, Europe, and India. The divergence is attributed to opposing trends in secondary organic aerosol(SOA) and sulfate coatings, driven by regional changes in the emission ratios of corresponding coating precursors to BC(volatile organic compounds-VOCs/BC and SO_(2)/BC). Projections under carbon neutrality reveal that R_(BC) will increase globally by 47%(118%) in 2050(2100), with strong convergent increases expected across major source regions. The R_(BC) increase, primarily driven by enhanced SOA coatings due to sharper BC reductions relative to VOCs, will enhance the global BC mass absorption cross-section(MAC) by 11%(17%) in 2050(2100).Consequently, although the global BC burden will decline sharply by 60%(76%), the enhanced MAC partially offsets the magnitude of the decline in the BC direct radiative effect, resulting in the moderation of global BC DRE decreases to 88%(92%) of the BC burden reductions in 2050(2100). This study highlights the globally enhanced BC aging and light absorption capacity under carbon neutrality, thereby partly offsetting the impact of BC direct emission reductions on future changes in BC radiative effects globally.展开更多
This study presents an interpretable surrogate framework for predicting pedestrian-leg injury severity that integrates high-fidelity finite-element(FE)simulations with a TabNet-based deep-learning model.We generated a...This study presents an interpretable surrogate framework for predicting pedestrian-leg injury severity that integrates high-fidelity finite-element(FE)simulations with a TabNet-based deep-learning model.We generated a parametric dataset of 3000 impact scenarios-covering ten vehicle types and various legform impactors-using automated FE runs configured via Latin hypercube sampling.After preprocessing and one-hot encoding of categorical features,we trained TabNet alongside Support-Vector Regression,Random Forest,and Decision-Tree ensembles.All models underwent hyperparameter tuning via Optuna’s Bayesian optimization coupled with repeated four-fold crossvalidation(20 trials per model).TabNet achieved the best balance of explanatory power and predictive accuracy,with an average R^(2)=0.94±0.01 and RMSE=0.14±0.02.On an independent test set,85%,88%,and 90%of predictions for tibial acceleration,knee-flexion angle,and shear displacement,respectively,fell within±20%of true peaks.SHAPbased analyses confirm that collision-point location and bumper geometry dominate injury outcomes.These results demonstrate TabNet’s capacity to deliver rapid,robust,and explainable injury predictions,offering actionable design insights for vehicle front-end optimization and regulatory assessment in early development stages.展开更多
Underwater pipeline inspection plays a vital role in the proactive maintenance and management of critical marine infrastructure and subaquatic systems.However,the inspection of underwater pipelines presents a challeng...Underwater pipeline inspection plays a vital role in the proactive maintenance and management of critical marine infrastructure and subaquatic systems.However,the inspection of underwater pipelines presents a challenge due to factors such as light scattering,absorption,restricted visibility,and ambient noise.The advancement of deep learning has introduced powerful techniques for processing large amounts of unstructured and imperfect data collected from underwater environments.This study evaluated the efficacy of the You Only Look Once(YOLO)algorithm,a real-time object detection and localization model based on convolutional neural networks,in identifying and classifying various types of pipeline defects in underwater settings.YOLOv8,the latest evolution in the YOLO family,integrates advanced capabilities,such as anchor-free detection,a cross-stage partial network backbone for efficient feature extraction,and a feature pyramid network+path aggregation network neck for robust multi-scale object detection,which make it particularly well-suited for complex underwater environments.Due to the lack of suitable open-access datasets for underwater pipeline defects,a custom dataset was captured using a remotely operated vehicle in a controlled environment.This application has the following assets available for use.Extensive experimentation demonstrated that YOLOv8 X-Large consistently outperformed other models in terms of pipe defect detection and classification and achieved a strong balance between precision and recall in identifying pipeline cracks,rust,corners,defective welds,flanges,tapes,and holes.This research establishes the baseline performance of YOLOv8 for underwater defect detection and showcases its potential to enhance the reliability and efficiency of pipeline inspection tasks in challenging underwater environments.展开更多
Background Efficient disaster victim detection(DVD)in urban areas after natural disasters is crucial for minimizing losses.However,conventional search and rescue(SAR)methods often experience delays,which can hinder th...Background Efficient disaster victim detection(DVD)in urban areas after natural disasters is crucial for minimizing losses.However,conventional search and rescue(SAR)methods often experience delays,which can hinder the timely detection of victims.SAR teams face various challenges,including limited access to debris and collapsed structures,safety risks due to unstable conditions,and disrupted communication networks.Methods In this paper,we present DeepSafe,a novel two-level deep learning approach for multilevel classification and object detection using a simulated disaster victim dataset.DeepSafe first employs YOLOv8 to classify images into victim and non-victim categories.Subsequently,Detectron2 is used to precisely locate and outline the victims.Results Experimental results demonstrate the promising performance of DeepSafe in both victim classification and detection.The model effectively identified and located victims under the challenging conditions presented in the dataset.Conclusion DeepSafe offers a practical tool for real-time disaster management and SAR operations,significantly improving conventional methods by reducing delays and enhancing victim detection accuracy in disaster-stricken urban areas.展开更多
Cyclohexene is an important raw material in the production of nylon.Selective hydrogenation of benzene is a key method for preparing cyclohexene.However,the Ru catalysts used in current industrial processes still face...Cyclohexene is an important raw material in the production of nylon.Selective hydrogenation of benzene is a key method for preparing cyclohexene.However,the Ru catalysts used in current industrial processes still face challenges,including high metal usage,high process costs,and low cyclohexene yield.This study utilizes existing literature data combined with machine learning methods to analyze the factors influencing benzene conversion,cyclohexene selectivity,and yield in the benzene hydrogenation to cyclohexene reaction.It constructs predictive models based on XGBoost and Random Forest algorithms.After analysis,it was found that reaction time,Ru content,and space velocity are key factors influencing cyclohexene yield,selectivity,and benzene conversion.Shapley Additive Explanations(SHAP)analysis and feature importance analysis further revealed the contribution of each variable to the reaction outcomes.Additionally,we randomly generated one million variable combinations using the Dirichlet distribution to attempt to predict high-yield catalyst formulations.This paper provides new insights into the application of machine learning in heterogeneous catalysis and offers some reference for further research.展开更多
The electrocatalytic reduction of nitric oxide for ammonia synthesis(NORR)is a key green energy conversion technology.Its efficiency relies on high-performance electrocatalysts to enhance both ammonia yield(Y_(NH3))an...The electrocatalytic reduction of nitric oxide for ammonia synthesis(NORR)is a key green energy conversion technology.Its efficiency relies on high-performance electrocatalysts to enhance both ammonia yield(Y_(NH3))and Faradaic efficiency(F_(NH3)).However,conventional experimental methods for screening high-activity NORR catalysts often entail high resource consumption and time costs.Machine learning combined with SHAP feature analysis was employed to establish a stacked ensemble model that integrates multiple algorithms,to allow for a systematic investigation of the key descriptors governing NORR performance based on an experimental dataset.Evaluation of eight model algorithms revealed that the Stacked-SVR model achieved an R^(2)of 0.9223 and an RMSE of 0.0608 for predicting on the test set,whereas the Stacked-RF model achieved an R^(2)of 0.9042 and an RMSE of 0.0900 for predicting.The stacked ensemble model integrates the strengths of individual algorithms and demonstrates strong NORR prediction performance while avoiding overfitting.SHAP feature analysis results revealed that the Cu content in the catalyst composition has the most significant impact on catalytic performance.Moreover,the combination of the wet chemical reduction synthesis,a carbon fiber(CF)conductive substrate,and HCl electrolyte is more favorable for enhancing catalytic activity.Additionally,moderately lowering the working potential,controlling the electrolyte volume at low to medium levels,reducing catalyst loading,and increasing electrolyte concentration were found to synergistically enhance both and.展开更多
Federated Learning(FL)has become a leading decentralized solution that enables multiple clients to train a model in a collaborative environment without directly sharing raw data,making it suitable for privacy-sensitiv...Federated Learning(FL)has become a leading decentralized solution that enables multiple clients to train a model in a collaborative environment without directly sharing raw data,making it suitable for privacy-sensitive applications such as healthcare,finance,and smart systems.As the field continues to evolve,the research field has become more complex and scattered,covering different system designs,training methods,and privacy techniques.This survey is organized around the three core challenges:how the data is distributed,how models are synchronized,and how to defend against attacks.It provides a structured and up-to-date review of FL research from 2023 to 2025,offering a unified taxonomy that categorizes works by data distribution(Horizontal FL,Vertical FL,Federated Transfer Learning,and Personalized FL),training synchronization(synchronous and asynchronous FL),optimization strategies,and threat models(data leakage and poisoning attacks).In particular,we summarize the latest contributions in Vertical FL frameworks for secure multi-party learning,communication-efficient Horizontal FL,and domain-adaptive Federated Transfer Learning.Furthermore,we examine synchronization techniques addressing system heterogeneity,including straggler mitigation in synchronous FL and staleness management in asynchronous FL.The survey covers security threats in FL,such as gradient inversion,membership inference,and poisoning attacks,as well as their defense strategies that include privacy-preserving aggregation and anomaly detection.The paper concludes by outlining unresolved issues and highlighting challenges in handling personalized models,scalability,and real-world adoption.展开更多
Ever since the research in machine learning gained traction in recent years,it has been employed to address challenges in a wide variety of domains,including mechanical devices.Most of the machine learning models are ...Ever since the research in machine learning gained traction in recent years,it has been employed to address challenges in a wide variety of domains,including mechanical devices.Most of the machine learning models are built on the assumption of a static learning environment,but in practical situations,the data generated by the process is dynamic.This evolution of the data is termed concept drift.This research paper presents an approach for predictingmechanical failure in real-time using incremental learning based on the statistically calculated parameters of mechanical equipment.The method proposed here is applicable to allmechanical devices that are susceptible to failure or operational degradation.The proposed method in this paper is equipped with the capacity to detect the drift in data generation and adaptation.The proposed approach evaluates the machine learning and deep learning models for their efficacy in handling the errors related to industrial machines due to their dynamic nature.It is observed that,in the settings without concept drift in the data,methods like SVM and Random Forest performed better compared to deep neural networks.However,this resulted in poor sensitivity for the smallest drift in the machine data reported as a drift.In this perspective,DNN generated the stable drift detection method;it reported an accuracy of 84%and an AUC of 0.87 while detecting only a single drift point,indicating the stability to performbetter in detecting and adapting to new data in the drifting environments under industrial measurement settings.展开更多
Split Learning(SL)has been promoted as a promising collaborative machine learning technique designed to address data privacy and resource efficiency.Specifically,neural networks are divided into client and server subn...Split Learning(SL)has been promoted as a promising collaborative machine learning technique designed to address data privacy and resource efficiency.Specifically,neural networks are divided into client and server subnetworks in order to mitigate the exposure of sensitive data and reduce the overhead on client devices,thereby making SL particularly suitable for resource-constrained devices.Although SL prevents the direct transmission of raw data,it does not alleviate entirely the risk of privacy breaches.In fact,the data intermediately transmitted to the server sub-model may include patterns or information that could reveal sensitive data.Moreover,achieving a balance between model utility and data privacy has emerged as a challenging problem.In this article,we propose a novel defense approach that combines:(i)Adversarial learning,and(ii)Network channel pruning.In particular,the proposed adversarial learning approach is specifically designed to reduce the risk of private data exposure while maintaining high performance for the utility task.On the other hand,the suggested channel pruning enables the model to adaptively adjust and reactivate pruned channels while conducting adversarial training.The integration of these two techniques reduces the informativeness of the intermediate data transmitted by the client sub-model,thereby enhancing its robustness against attribute inference attacks without adding significant computational overhead,making it wellsuited for IoT devices,mobile platforms,and Internet of Vehicles(IoV)scenarios.The proposed defense approach was evaluated using EfficientNet-B0,a widely adopted compact model,along with three benchmark datasets.The obtained results showcased its superior defense capability against attribute inference attacks compared to existing state-of-the-art methods.This research’s findings demonstrated the effectiveness of the proposed channel pruning-based adversarial training approach in achieving the intended compromise between utility and privacy within SL frameworks.In fact,the classification accuracy attained by the attackers witnessed a drastic decrease of 70%.展开更多
The Internet of Vehicles,or IoV,is expected to lessen pollution,ease traffic,and increase road safety.IoV entities’interconnectedness,however,raises the possibility of cyberattacks,which can have detrimental effects....The Internet of Vehicles,or IoV,is expected to lessen pollution,ease traffic,and increase road safety.IoV entities’interconnectedness,however,raises the possibility of cyberattacks,which can have detrimental effects.IoV systems typically send massive volumes of raw data to central servers,which may raise privacy issues.Additionally,model training on IoV devices with limited resources normally leads to slower training times and reduced service quality.We discuss a privacy-preserving Federated Split Learning with Tiny Machine Learning(TinyML)approach,which operates on IoV edge devices without sharing sensitive raw data.Specifically,we focus on integrating split learning(SL)with federated learning(FL)and TinyML models.FL is a decentralisedmachine learning(ML)technique that enables numerous edge devices to train a standard model while retaining data locally collectively.The article intends to thoroughly discuss the architecture and challenges associated with the increasing prevalence of SL in the IoV domain,coupled with FL and TinyML.The approach starts with the IoV learning framework,which includes edge computing,FL,SL,and TinyML,and then proceeds to discuss how these technologies might be integrated.We elucidate the comprehensive operational principles of Federated and split learning by examining and addressingmany challenges.We subsequently examine the integration of SL with FL and various applications of TinyML.Finally,exploring the potential integration of FL and SL with TinyML in the IoV domain is referred to as FSL-TM.It is a superior method for preserving privacy as it conducts model training on individual devices or edge nodes,thereby obviating the necessity for centralised data aggregation,which presents considerable privacy threats.The insights provided aim to help both researchers and practitioners understand the complicated terrain of FL and SL,hence facilitating advancement in this swiftly progressing domain.展开更多
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
基金supported in part by the National Natural Science Foundation of China(No.52177109)Key R&D Program of Hubei Province,China(No.2020BAB109).
文摘With the prevalence of renewable distributed energy resources(DERs)such as photovoltaics(PVs),modern active distribution networks(ADNs)suffer from voltage deviation and power quality issues.However,traditional voltage control methods often face a trade-off between efficiency and effectiveness,and rarely ensure robust voltage safety under typical state perturbations in practical distribution grids.In this paper,a robust model-free voltage regulation approach is proposed which simultaneously takes security and robustness into account.In this context,the voltage control problem is formulated as a constrained Markov decision process(CMDP).A safety-augmented multiagent deep deterministic policy gradient(MADDPG)algorithm is the trained to enable real-time collaborative optimization of ADNs,aiming to maintain nodal voltages within safe operational limits while minimizing total line losses.Moreover,a robust regulation loss is introduced to ensure reliable performance under various state perturbations in practical voltage controls.The proposed regulation algorithm effectively balance efficiency,safety,and robustness,and also demonstrates potential for generalizing these characteristics to other applications.Numerical studies vali-date the robustness of the proposed method under varying state perturbations on the IEEE test cases and the optimal integrated control performance when compared to other benchmarks.
基金appreciation to the Deanship of Research and Graduate Studies at King Khalid University for funding this work through the Large Group Project under grant number(RGP2/245/46)Princess Nourah bint Abdulrahman University Researchers Supporting Project number(PNURSP2026R333)Princess Nourah bint Abdulrahman University,Riyadh,Saudi Arabia.
文摘The expansion of 5G-enabled Internet of Things(IoT)networks,while enabling transformative applications,significantly increases the attack surface and necessitates security solutions that extend beyond traditional intrusion detection.Existing intrusion detection systems(IDSs)mainly operate in an open-loop manner,excelling at classification but lacking the ability for autonomous,safety-aware remediation.This gap is particularly critical in 5G environments,where manual intervention is too slow and naive automation can lead to severe service disruptions.To address this issue,we propose a novel Self-Healing Intrusion Detection System(SH-IDS)framework that develops a closed-loop cyber defense mechanism.The main technical contribution is the integration of a deep neural networkbased threat detector,which offers uncertainty-quantified predictions,with a safety-aware reinforcement learning(RL)engine formulated as a Constrained Markov Decision Process(CMDP).The CMDP explicitly models operational safety as cost constraints,and a new runtime safety shield actively adjusts any unsafe action proposed by the RL agent to the nearest safe alternative,ensuring operational integrity.Additionally,we introduce a composite utility function for the comprehensive evaluation of the system.Empirical analysis on the 5G-NIDD dataset demonstrates the superior performance of our framework:the detector achieves 98.26%accuracy,while the safe RL agent learns effective mitigation policies.Importantly,the safety shield blocked up to 70 unsafe actions under strict constraints,and analysis of the learned Q-tables confirms that the agent internalizes safety,avoiding overly disruptive actions,such as isolating nodes for minor threats.The system also maintains high efficiency with a compact model size of 121.7 KB and sub-millisecond latency,confirming its practical deployability for real-time 5G-IoT security.
基金supported by the National Natural Science Foundation of China(52007173 and U22B2098).
文摘The real-time AC optimal power flow(OPF)problem is a key issue in making fast and accurate decisions to ensure the safety and economy of power systems.With the rapid development of renewable energies,the fluctuation has grown more vibrant,thus a novel approach called safe deep reinforcement learning is proposed in this paper.Herein,the real-time ACOPF problem is modeled as a constrained Markov decision process,and primal-dual optimization(PDO)based proximal policy optimization(PPO)is used to learn the optimal generator outputs in the primal domain and security constraints in the dual domain,which avoids manually selecting a trade-off between penalties for constraint violations and rewards for the economy.Before training,behavior cloning clones the expert experience into the initial weights of neural networks.Moreover,multiprocessing training is utilized to accelerate the training speed.Case studies are conducted on the IEEE 118-bus system and the modified IEEE 118-bus system.Compared with other methods,the experimental results show that the proposed method can achieve security and near-optimal economic goals by fast calculating the real-time ACOPF problem.
文摘Grasping is one of the most fundamental operations in modern robotics applications.While deep rein-forcement learning(DRL)has demonstrated strong potential in robotics,there is too much emphasis on maximizing the cumulative reward in executing tasks,and the potential safety risks are often ignored.In this paper,an optimization method based on safe reinforcement learning(Safe RL)is proposed to address the robotic grasping problem under safety constraints.Specifically,considering the obstacle avoidance constraints of the system,the grasping problem of the manipulator is modeled as a Constrained Markov Decision Process(CMDP).The Lagrange multiplier and a dynamic weighted mechanism are introduced into the Proximal Policy Optimization(PPO)framework,leading to the development of the dynamic weighted Lagrange PPO(DWL-PPO)algorithm.The behavior of violating safety constraints is punished while the policy is optimized in this proposed method.In addition,the orientation control of the end-effector is included in the reward function,and a compound reward function adapted to changes in pose is designed.Ultimately,the efficacy and advantages of the suggested method are proved by extensive training and testing in the Pybullet simulator.The results of grasping experiments reveal that the recommended approach provides superior safety and efficiency compared with other advanced RL methods and achieves a good trade-off between model learning and risk aversion.
基金supported by the National Key Research and Development Program of China(No.2021YFF0500600)Natural Science Foundation of Henan Province(No.252300421176)+1 种基金National Natural Science Foundation of China(No.22478361 and No.22108256)Frontier Exploration Projects of Longmen Laboratory(No.LMQYTSKT021)。
文摘Batteries play a crucial role in the storage and application of sustainable energy,yet their inherent safety risks are non-negligible.Traditional monitoring methods often suffer from high costs,time consumption,and limited scalability,making it increasingly difficult to meet the evolving demands of modern society.In this context,recent advancements in machine learning technology have emerged as a promising solution for predicting and monitoring battery states,offering innovative approaches to battery management systems(BMS).By transforming raw operational data into actionable insights,machine learning has shifted the paradigm from reactive to predictive battery safety management,significantly enhancing system reliability and risk mitigation capabilities.This review delves into the implementation of machine learning in battery state prediction,including dataset selection,feature extraction,and model training.It also highlights the latest progress of these models in key applications such as state of health(SOH),state of charge(SOC),thermal runaway warning,fault detection,and remaining useful life(RUL).Finally,we critically examined the challenges and opportunities associated with leveraging machine learning to improve battery safety and performance,providing a comprehensive perspective for future research in this rapidly advancing field.
文摘Low visibility conditions,particularly those caused by fog,significantly affect road safety and reduce drivers’ability to see ahead clearly.The conventional approaches used to address this problem primarily rely on instrument-based and fixed-threshold-based theoretical frameworks,which face challenges in adaptability and demonstrate lower performance under varying environmental conditions.To overcome these challenges,we propose a real-time visibility estimation model that leverages roadside CCTV cameras to monitor and identify visibility levels under different weather conditions.The proposedmethod begins by identifying specific regions of interest(ROI)in the CCTVimages and focuses on extracting specific features such as the number of lines and contours detected within these regions.These features are then provided as an input to the proposed hierarchical clusteringmodel,which classifies them into different visibility levels without the need for predefined rules and threshold values.In the proposed approach,we used two different distance similaritymetrics,namely dynamic time warping(DTW)and Euclidean distance,alongside the proposed hierarchical clustering model and noted its performance in terms of numerous evaluation measures.The proposed model achieved an average accuracy of 97.81%,precision of 91.31%,recall of 91.25%,and F1-score of 91.27% using theDTWdistancemetric.We also conducted experiments for other deep learning(DL)-based models used in the literature and compared their performances with the proposed model.The experimental results demonstrate that the proposedmodel ismore adaptable and consistent compared to themethods used in the literature.The proposedmethod provides drivers real-time and accurate visibility information and enhances road safety during low visibility conditions.
基金Guangzhou Metro Scientific Research Project(No.JT204-100111-23001)Chongqing Municipal Special Project for Technological Innovation and Application Development(No.CSTB2022TIAD-KPX0101)Science and Technology Research and Development Program of China State Railway Group Co.,Ltd.(No.N2023G045)。
文摘The uplift resistance of the soil overlying shield tunnels significantly impacts their anti-floating stability.However,research on uplift resistance concerning special-shaped shield tunnels is limited.This study combines numerical simulation with machine learning techniques to explore this issue.It presents a summary of special-shaped tunnel geometries and introduces a shape coefficient.Through the finite element software,Plaxis3D,the study simulates six key parameters—shape coefficient,burial depth ratio,tunnel’s longest horizontal length,internal friction angle,cohesion,and soil submerged bulk density—that impact uplift resistance across different conditions.Employing XGBoost and ANN methods,the feature importance of each parameter was analyzed based on the numerical simulation results.The findings demonstrate that a tunnel shape more closely resembling a circle leads to reduced uplift resistance in the overlying soil,whereas other parameters exhibit the contrary effects.Furthermore,the study reveals a diminishing trend in the feature importance of buried depth ratio,internal friction angle,tunnel longest horizontal length,cohesion,soil submerged bulk density,and shape coefficient in influencing uplift resistance.
基金supported by the National Natural Science Foundation of China (42505149,41925023,U2342223,42105069,and 91744208)the China Postdoctoral Science Foundation (2025M770303)+1 种基金the Fundamental Research Funds for the Central Universities (14380230)the Jiangsu Funding Program for Excellent Postdoctoral Talent,and Jiangsu Collaborative Innovation Center of Climate Change。
文摘Countries around the world have been making efforts to reduce pollutant emissions. However, the response of global black carbon(BC) aging to emission changes remains unclear. Using the Community Atmosphere Model version 6 with a machine-learning-integrated four-mode version of the Modal Aerosol Module, we quantify global BC aging responses to emission reductions for 2011–2018 and for 2050 and 2100 under carbon neutrality. During 2011–18, global trends in BC aging degree(mass ratio of coatings to BC, R_(BC)) exhibited marked regional disparities, with a significant increase in China(5.4% yr^(-1)), which contrasts with minimal changes in the USA, Europe, and India. The divergence is attributed to opposing trends in secondary organic aerosol(SOA) and sulfate coatings, driven by regional changes in the emission ratios of corresponding coating precursors to BC(volatile organic compounds-VOCs/BC and SO_(2)/BC). Projections under carbon neutrality reveal that R_(BC) will increase globally by 47%(118%) in 2050(2100), with strong convergent increases expected across major source regions. The R_(BC) increase, primarily driven by enhanced SOA coatings due to sharper BC reductions relative to VOCs, will enhance the global BC mass absorption cross-section(MAC) by 11%(17%) in 2050(2100).Consequently, although the global BC burden will decline sharply by 60%(76%), the enhanced MAC partially offsets the magnitude of the decline in the BC direct radiative effect, resulting in the moderation of global BC DRE decreases to 88%(92%) of the BC burden reductions in 2050(2100). This study highlights the globally enhanced BC aging and light absorption capacity under carbon neutrality, thereby partly offsetting the impact of BC direct emission reductions on future changes in BC radiative effects globally.
基金sponsored by the National Natural Science Foundation of China(No.U21A20165,No.52072057).
文摘This study presents an interpretable surrogate framework for predicting pedestrian-leg injury severity that integrates high-fidelity finite-element(FE)simulations with a TabNet-based deep-learning model.We generated a parametric dataset of 3000 impact scenarios-covering ten vehicle types and various legform impactors-using automated FE runs configured via Latin hypercube sampling.After preprocessing and one-hot encoding of categorical features,we trained TabNet alongside Support-Vector Regression,Random Forest,and Decision-Tree ensembles.All models underwent hyperparameter tuning via Optuna’s Bayesian optimization coupled with repeated four-fold crossvalidation(20 trials per model).TabNet achieved the best balance of explanatory power and predictive accuracy,with an average R^(2)=0.94±0.01 and RMSE=0.14±0.02.On an independent test set,85%,88%,and 90%of predictions for tibial acceleration,knee-flexion angle,and shear displacement,respectively,fell within±20%of true peaks.SHAPbased analyses confirm that collision-point location and bumper geometry dominate injury outcomes.These results demonstrate TabNet’s capacity to deliver rapid,robust,and explainable injury predictions,offering actionable design insights for vehicle front-end optimization and regulatory assessment in early development stages.
文摘Underwater pipeline inspection plays a vital role in the proactive maintenance and management of critical marine infrastructure and subaquatic systems.However,the inspection of underwater pipelines presents a challenge due to factors such as light scattering,absorption,restricted visibility,and ambient noise.The advancement of deep learning has introduced powerful techniques for processing large amounts of unstructured and imperfect data collected from underwater environments.This study evaluated the efficacy of the You Only Look Once(YOLO)algorithm,a real-time object detection and localization model based on convolutional neural networks,in identifying and classifying various types of pipeline defects in underwater settings.YOLOv8,the latest evolution in the YOLO family,integrates advanced capabilities,such as anchor-free detection,a cross-stage partial network backbone for efficient feature extraction,and a feature pyramid network+path aggregation network neck for robust multi-scale object detection,which make it particularly well-suited for complex underwater environments.Due to the lack of suitable open-access datasets for underwater pipeline defects,a custom dataset was captured using a remotely operated vehicle in a controlled environment.This application has the following assets available for use.Extensive experimentation demonstrated that YOLOv8 X-Large consistently outperformed other models in terms of pipe defect detection and classification and achieved a strong balance between precision and recall in identifying pipeline cracks,rust,corners,defective welds,flanges,tapes,and holes.This research establishes the baseline performance of YOLOv8 for underwater defect detection and showcases its potential to enhance the reliability and efficiency of pipeline inspection tasks in challenging underwater environments.
基金Supported by European Union’s Horizon 2020 Research and Innovation Program(739578)the Government of the Republic of Cyprus through the Deputy Ministry of Research,Innovation,and Digital Policy.
文摘Background Efficient disaster victim detection(DVD)in urban areas after natural disasters is crucial for minimizing losses.However,conventional search and rescue(SAR)methods often experience delays,which can hinder the timely detection of victims.SAR teams face various challenges,including limited access to debris and collapsed structures,safety risks due to unstable conditions,and disrupted communication networks.Methods In this paper,we present DeepSafe,a novel two-level deep learning approach for multilevel classification and object detection using a simulated disaster victim dataset.DeepSafe first employs YOLOv8 to classify images into victim and non-victim categories.Subsequently,Detectron2 is used to precisely locate and outline the victims.Results Experimental results demonstrate the promising performance of DeepSafe in both victim classification and detection.The model effectively identified and located victims under the challenging conditions presented in the dataset.Conclusion DeepSafe offers a practical tool for real-time disaster management and SAR operations,significantly improving conventional methods by reducing delays and enhancing victim detection accuracy in disaster-stricken urban areas.
基金Supported by CAS Basic and Interdisciplinary Frontier Scientific Research Pilot Project(XDB1190300,XDB1190302)Youth Innovation Promotion Association CAS(Y2021056)+1 种基金Joint Fund of the Yulin University and the Dalian National Laboratory for Clean Energy(YLU-DNL Fund 2022007)The special fund for Science and Technology Innovation Teams of Shanxi Province(202304051001007)。
文摘Cyclohexene is an important raw material in the production of nylon.Selective hydrogenation of benzene is a key method for preparing cyclohexene.However,the Ru catalysts used in current industrial processes still face challenges,including high metal usage,high process costs,and low cyclohexene yield.This study utilizes existing literature data combined with machine learning methods to analyze the factors influencing benzene conversion,cyclohexene selectivity,and yield in the benzene hydrogenation to cyclohexene reaction.It constructs predictive models based on XGBoost and Random Forest algorithms.After analysis,it was found that reaction time,Ru content,and space velocity are key factors influencing cyclohexene yield,selectivity,and benzene conversion.Shapley Additive Explanations(SHAP)analysis and feature importance analysis further revealed the contribution of each variable to the reaction outcomes.Additionally,we randomly generated one million variable combinations using the Dirichlet distribution to attempt to predict high-yield catalyst formulations.This paper provides new insights into the application of machine learning in heterogeneous catalysis and offers some reference for further research.
文摘The electrocatalytic reduction of nitric oxide for ammonia synthesis(NORR)is a key green energy conversion technology.Its efficiency relies on high-performance electrocatalysts to enhance both ammonia yield(Y_(NH3))and Faradaic efficiency(F_(NH3)).However,conventional experimental methods for screening high-activity NORR catalysts often entail high resource consumption and time costs.Machine learning combined with SHAP feature analysis was employed to establish a stacked ensemble model that integrates multiple algorithms,to allow for a systematic investigation of the key descriptors governing NORR performance based on an experimental dataset.Evaluation of eight model algorithms revealed that the Stacked-SVR model achieved an R^(2)of 0.9223 and an RMSE of 0.0608 for predicting on the test set,whereas the Stacked-RF model achieved an R^(2)of 0.9042 and an RMSE of 0.0900 for predicting.The stacked ensemble model integrates the strengths of individual algorithms and demonstrates strong NORR prediction performance while avoiding overfitting.SHAP feature analysis results revealed that the Cu content in the catalyst composition has the most significant impact on catalytic performance.Moreover,the combination of the wet chemical reduction synthesis,a carbon fiber(CF)conductive substrate,and HCl electrolyte is more favorable for enhancing catalytic activity.Additionally,moderately lowering the working potential,controlling the electrolyte volume at low to medium levels,reducing catalyst loading,and increasing electrolyte concentration were found to synergistically enhance both and.
文摘Federated Learning(FL)has become a leading decentralized solution that enables multiple clients to train a model in a collaborative environment without directly sharing raw data,making it suitable for privacy-sensitive applications such as healthcare,finance,and smart systems.As the field continues to evolve,the research field has become more complex and scattered,covering different system designs,training methods,and privacy techniques.This survey is organized around the three core challenges:how the data is distributed,how models are synchronized,and how to defend against attacks.It provides a structured and up-to-date review of FL research from 2023 to 2025,offering a unified taxonomy that categorizes works by data distribution(Horizontal FL,Vertical FL,Federated Transfer Learning,and Personalized FL),training synchronization(synchronous and asynchronous FL),optimization strategies,and threat models(data leakage and poisoning attacks).In particular,we summarize the latest contributions in Vertical FL frameworks for secure multi-party learning,communication-efficient Horizontal FL,and domain-adaptive Federated Transfer Learning.Furthermore,we examine synchronization techniques addressing system heterogeneity,including straggler mitigation in synchronous FL and staleness management in asynchronous FL.The survey covers security threats in FL,such as gradient inversion,membership inference,and poisoning attacks,as well as their defense strategies that include privacy-preserving aggregation and anomaly detection.The paper concludes by outlining unresolved issues and highlighting challenges in handling personalized models,scalability,and real-world adoption.
文摘Ever since the research in machine learning gained traction in recent years,it has been employed to address challenges in a wide variety of domains,including mechanical devices.Most of the machine learning models are built on the assumption of a static learning environment,but in practical situations,the data generated by the process is dynamic.This evolution of the data is termed concept drift.This research paper presents an approach for predictingmechanical failure in real-time using incremental learning based on the statistically calculated parameters of mechanical equipment.The method proposed here is applicable to allmechanical devices that are susceptible to failure or operational degradation.The proposed method in this paper is equipped with the capacity to detect the drift in data generation and adaptation.The proposed approach evaluates the machine learning and deep learning models for their efficacy in handling the errors related to industrial machines due to their dynamic nature.It is observed that,in the settings without concept drift in the data,methods like SVM and Random Forest performed better compared to deep neural networks.However,this resulted in poor sensitivity for the smallest drift in the machine data reported as a drift.In this perspective,DNN generated the stable drift detection method;it reported an accuracy of 84%and an AUC of 0.87 while detecting only a single drift point,indicating the stability to performbetter in detecting and adapting to new data in the drifting environments under industrial measurement settings.
基金supported by a grant(No.CRPG-25-2054)under the Cybersecurity Research and Innovation Pioneers Initiative,provided by the National Cybersecurity Authority(NCA)in the Kingdom of Saudi Arabia.
文摘Split Learning(SL)has been promoted as a promising collaborative machine learning technique designed to address data privacy and resource efficiency.Specifically,neural networks are divided into client and server subnetworks in order to mitigate the exposure of sensitive data and reduce the overhead on client devices,thereby making SL particularly suitable for resource-constrained devices.Although SL prevents the direct transmission of raw data,it does not alleviate entirely the risk of privacy breaches.In fact,the data intermediately transmitted to the server sub-model may include patterns or information that could reveal sensitive data.Moreover,achieving a balance between model utility and data privacy has emerged as a challenging problem.In this article,we propose a novel defense approach that combines:(i)Adversarial learning,and(ii)Network channel pruning.In particular,the proposed adversarial learning approach is specifically designed to reduce the risk of private data exposure while maintaining high performance for the utility task.On the other hand,the suggested channel pruning enables the model to adaptively adjust and reactivate pruned channels while conducting adversarial training.The integration of these two techniques reduces the informativeness of the intermediate data transmitted by the client sub-model,thereby enhancing its robustness against attribute inference attacks without adding significant computational overhead,making it wellsuited for IoT devices,mobile platforms,and Internet of Vehicles(IoV)scenarios.The proposed defense approach was evaluated using EfficientNet-B0,a widely adopted compact model,along with three benchmark datasets.The obtained results showcased its superior defense capability against attribute inference attacks compared to existing state-of-the-art methods.This research’s findings demonstrated the effectiveness of the proposed channel pruning-based adversarial training approach in achieving the intended compromise between utility and privacy within SL frameworks.In fact,the classification accuracy attained by the attackers witnessed a drastic decrease of 70%.
文摘The Internet of Vehicles,or IoV,is expected to lessen pollution,ease traffic,and increase road safety.IoV entities’interconnectedness,however,raises the possibility of cyberattacks,which can have detrimental effects.IoV systems typically send massive volumes of raw data to central servers,which may raise privacy issues.Additionally,model training on IoV devices with limited resources normally leads to slower training times and reduced service quality.We discuss a privacy-preserving Federated Split Learning with Tiny Machine Learning(TinyML)approach,which operates on IoV edge devices without sharing sensitive raw data.Specifically,we focus on integrating split learning(SL)with federated learning(FL)and TinyML models.FL is a decentralisedmachine learning(ML)technique that enables numerous edge devices to train a standard model while retaining data locally collectively.The article intends to thoroughly discuss the architecture and challenges associated with the increasing prevalence of SL in the IoV domain,coupled with FL and TinyML.The approach starts with the IoV learning framework,which includes edge computing,FL,SL,and TinyML,and then proceeds to discuss how these technologies might be integrated.We elucidate the comprehensive operational principles of Federated and split learning by examining and addressingmany challenges.We subsequently examine the integration of SL with FL and various applications of TinyML.Finally,exploring the potential integration of FL and SL with TinyML in the IoV domain is referred to as FSL-TM.It is a superior method for preserving privacy as it conducts model training on individual devices or edge nodes,thereby obviating the necessity for centralised data aggregation,which presents considerable privacy threats.The insights provided aim to help both researchers and practitioners understand the complicated terrain of FL and SL,hence facilitating advancement in this swiftly progressing domain.
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.
