Reconfigurable intelligent surface(RIS)have been cast as a promising alternative to alleviate blockage vulnerability and enhance coverage capability for terahertz(THz)communications.Owing to large-scale array elements...Reconfigurable intelligent surface(RIS)have been cast as a promising alternative to alleviate blockage vulnerability and enhance coverage capability for terahertz(THz)communications.Owing to large-scale array elements at transceivers and RIS,the codebook based beamforming can be utilized in a computationally efficient manner.However,the codeword selection for analog beamforming is an intractable combinatorial optimization(CO)problem.To this end,by taking the CO problem as a classification problem,a multi-task learning based analog beam selection(MTL-ABS)framework is developed to implement cooperative beam selection concurrently at transceivers and RIS.In addition,residual network and self-attention mechanism are used to combat the network degradation and mine intrinsic THz channel features.Finally,the network convergence is analyzed from a blockwise perspective,and numerical results demonstrate that the MTL-ABS framework greatly decreases the beam selection overhead and achieves near optimal sum-rate compared with heuristic search based counterparts.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
Underground engineering projects such as deep tunnel excavation often encounter rockburst disasters accompanied by numerous microseismic events.Rapid interpretation of microseismic signals is crucial for the timely id...Underground engineering projects such as deep tunnel excavation often encounter rockburst disasters accompanied by numerous microseismic events.Rapid interpretation of microseismic signals is crucial for the timely identification of rockbursts.However,conventional processing encompasses multi-step workflows,including classification,denoising,picking,locating,and computational analysis,coupled with manual intervention,which collectively compromise the reliability of early warnings.To address these challenges,this study innovatively proposes the“microseismic stethoscope"-a multi-task machine learning and deep learning model designed for the automated processing of massive microseismic signals.This model efficiently extracts three key parameters that are necessary for recognizing rockburst disasters:rupture location,microseismic energy,and moment magnitude.Specifically,the model extracts raw waveform features from three dedicated sub-networks:a classifier for source zone classification,and two regressors for microseismic energy and moment magnitude estimation.This model demonstrates superior efficiency compared to traditional processing and semi-automated processing,reducing per-event processing time from 0.71 s to 0.49 s to merely 0.036 s.It concurrently achieves 98%accuracy in source zone classification,with microseismic energy and moment magnitude estimation errors of 0.13 and 0.05,respectively.This model has been well applied and validated in the Daxiagu Tunnel case in Sichuan,China.The application results indicate that the model is as accurate as traditional methods in determining source parameters,and thus can be used to identify potential geomechanical processes of rockburst disasters.By enhancing the signal processing reliability of microseismic events,the proposed model in this study presents a significant advancement in the identification of rockburst disasters.展开更多
With the proliferation of Internet of Things(IoT)devices,securing these interconnected systems against cyberattacks has become a critical challenge.Traditional security paradigms often fail to cope with the scale and ...With the proliferation of Internet of Things(IoT)devices,securing these interconnected systems against cyberattacks has become a critical challenge.Traditional security paradigms often fail to cope with the scale and diversity of IoT network traffic.This paper presents a comparative benchmark of classic machine learning(ML)and state-of-the-art deep learning(DL)algorithms for IoT intrusion detection.Our methodology employs a twophased approach:a preliminary pilot study using a custom-generated dataset to establish baselines,followed by a comprehensive evaluation on the large-scale CICIoTDataset2023.We benchmarked algorithms including Random Forest,XGBoost,CNN,and StackedLSTM.The results indicate that while top-performingmodels frombothcategories achieve over 99%classification accuracy,this metric masks a crucial performance trade-off.We demonstrate that treebased ML ensembles exhibit superior precision(91%)in identifying benign traffic,making them effective at reducing false positives.Conversely,DL models demonstrate superior recall(96%),making them better suited for minimizing the interruption of legitimate traffic.We conclude that the selection of an optimal model is not merely a matter of maximizing accuracy but is a strategic choice dependent on the specific security priority either minimizing false alarms or ensuring service availability.Thiswork provides a practical framework for deploying context-aware security solutions in diverse IoT environments.展开更多
With the growing complexity and decentralization of network systems,the attack surface has expanded,which has led to greater concerns over network threats.In this context,artificial intelligence(AI)-based network intr...With the growing complexity and decentralization of network systems,the attack surface has expanded,which has led to greater concerns over network threats.In this context,artificial intelligence(AI)-based network intrusion detection systems(NIDS)have been extensively studied,and recent efforts have shifted toward integrating distributed learning to enable intelligent and scalable detection mechanisms.However,most existing works focus on individual distributed learning frameworks,and there is a lack of systematic evaluations that compare different algorithms under consistent conditions.In this paper,we present a comprehensive evaluation of representative distributed learning frameworks—Federated Learning(FL),Split Learning(SL),hybrid collaborative learning(SFL),and fully distributed learning—in the context of AI-driven NIDS.Using recent benchmark intrusion detection datasets,a unified model backbone,and controlled distributed scenarios,we assess these frameworks across multiple criteria,including detection performance,communication cost,computational efficiency,and convergence behavior.Our findings highlight distinct trade-offs among the distributed learning frameworks,demonstrating that the optimal choice depends strongly on systemconstraints such as bandwidth availability,node resources,and data distribution.This work provides the first holistic analysis of distributed learning approaches for AI-driven NIDS and offers practical guidelines for designing secure and efficient intrusion detection systems in decentralized environments.展开更多
Convolutional neural networks(CNNs)have shown remarkable success across numerous tasks such as image classification,yet the theoretical understanding of their convergence remains underdeveloped compared to their empir...Convolutional neural networks(CNNs)have shown remarkable success across numerous tasks such as image classification,yet the theoretical understanding of their convergence remains underdeveloped compared to their empirical achievements.In this paper,the first filter learning framework with convergence-guaranteed learning laws for end-to-end learning of deep CNNs is proposed.Novel update laws with convergence analysis are formulated based on the mathematical representation of each layer in convolutional neural networks.The proposed learning laws enable concurrent updates of weights across all layers of the deep convolutional neural network and the analysis shows that the training errors converge to certain bounds which are dependent on the approximation errors.Case studies are conducted on benchmark datasets and the results show that the proposed concurrent filter learning framework guarantees the convergence and offers more consistent and reliable results during training with a trade-off in performance compared to stochastic gradient descent methods.This framework represents a significant step towards enhancing the reliability and effectiveness of deep convolutional neural network by developing a theoretical analysis which allows practical implementation of the learning laws with automatic tuning of the learning rate to guarantee the convergence during training.展开更多
Distributed Denial-of-Service(DDoS)attacks pose severe threats to Industrial Control Networks(ICNs),where service disruption can cause significant economic losses and operational risks.Existing signature-based methods...Distributed Denial-of-Service(DDoS)attacks pose severe threats to Industrial Control Networks(ICNs),where service disruption can cause significant economic losses and operational risks.Existing signature-based methods are ineffective against novel attacks,and traditional machine learning models struggle to capture the complex temporal dependencies and dynamic traffic patterns inherent in ICN environments.To address these challenges,this study proposes a deep feature-driven hybrid framework that integrates Transformer,BiLSTM,and KNN to achieve accurate and robust DDoS detection.The Transformer component extracts global temporal dependencies from network traffic flows,while BiLSTM captures fine-grained sequential dynamics.The learned embeddings are then classified using an instance-based KNN layer,enhancing decision boundary precision.This cascaded architecture balances feature abstraction and locality preservation,improving both generalization and robustness.The proposed approach was evaluated on a newly collected real-time ICN traffic dataset and further validated using the public CIC-IDS2017 and Edge-IIoT datasets to demonstrate generalization.Comprehensive metrics including accuracy,precision,recall,F1-score,ROC-AUC,PR-AUC,false positive rate(FPR),and detection latency were employed.Results show that the hybrid framework achieves 98.42%accuracy with an ROC-AUC of 0.992 and FPR below 1%,outperforming baseline machine learning and deep learning models.Robustness experiments under Gaussian noise perturbations confirmed stable performance with less than 2%accuracy degradation.Moreover,detection latency remained below 2.1 ms per sample,indicating suitability for real-time ICS deployment.In summary,the proposed hybrid temporal learning and instance-based classification model offers a scalable and effective solution for DDoS detection in industrial control environments.By combining global contextual modeling,sequential learning,and instance-based refinement,the framework demonstrates strong adaptability across datasets and resilience against noise,providing practical utility for safeguarding critical infrastructure.展开更多
As the types of traffic requests increase,the elastic optical network(EON)is considered as a promising architecture to carry multiple types of traffic requests simultaneously,including immediate reservation(IR)and adv...As the types of traffic requests increase,the elastic optical network(EON)is considered as a promising architecture to carry multiple types of traffic requests simultaneously,including immediate reservation(IR)and advance reservation(AR).Various resource allocation schemes for IR/AR requests have been designed in EON to reduce bandwidth blocking probability(BBP).However,these schemes do not consider different transmission requirements of IR requests and cannot maintain a low BBP for high-priority requests.In this paper,multi-priority is considered in the hybrid IR/AR request scenario.We modify the asynchronous advantage actor critic(A3C)model and propose an A3C-assisted priority resource allocation(APRA)algorithm.The APRA integrates priority and transmission quality of IR requests to design the A3C reward function,then dynamically allocates dedicated resources for different IR requests according to the time-varying requirements.By maximizing the reward,the transmission quality of IR requests can be matched with the priority,and lower BBP for high-priority IR requests can be ensured.Simulation results show that the APRA reduces the BBP of high-priority IR requests from 0.0341 to0.0138,and the overall network operation gain is improved by 883 compared to the scheme without considering the priority.展开更多
Wi-Fi technology has evolved significantly since its introduction in 1997,advancing to Wi-Fi 6 as the latest standard,with Wi-Fi 7 currently under development.Despite these advancements,integrating machine learning in...Wi-Fi technology has evolved significantly since its introduction in 1997,advancing to Wi-Fi 6 as the latest standard,with Wi-Fi 7 currently under development.Despite these advancements,integrating machine learning into Wi-Fi networks remains challenging,especially in decentralized environments with multiple access points(mAPs).This paper is a short review that summarizes the potential applications of federated reinforcement learning(FRL)across eight key areas of Wi-Fi functionality,including channel access,link adaptation,beamforming,multi-user transmissions,channel bonding,multi-link operation,spatial reuse,and multi-basic servic set(multi-BSS)coordination.FRL is highlighted as a promising framework for enabling decentralized training and decision-making while preserving data privacy.To illustrate its role in practice,we present a case study on link activation in a multi-link operation(MLO)environment with multiple APs.Through theoretical discussion and simulation results,the study demonstrates how FRL can improve performance and reliability,paving the way for more adaptive and collaborative Wi-Fi networks in the era of Wi-Fi 7 and beyond.展开更多
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
In response to the rising demand for low-latency,computation-intensive applications in vehicular networks,this paper proposes an adaptive task offloading approach for Vehicle-to-Everything(V2X)environments.Leveraging ...In response to the rising demand for low-latency,computation-intensive applications in vehicular networks,this paper proposes an adaptive task offloading approach for Vehicle-to-Everything(V2X)environments.Leveraging an enhanced Multi-Agent Deep Deterministic Policy Gradient(MADDPG)algorithm with an attention mechanism,the proposed approach optimizes computation offloading and resource allocation,aiming to minimize energy consumption and service delay.In this paper,vehicles dynamically offload computing-intensive tasks to both nearby vehicles through V2V links and roadside units through V2I links.The adaptive attention mechanism enables the system to prioritize relevant state information,leading to faster convergence.Simulations conducted in a realistic urban V2X scenario demonstrate that the proposed Attention-enhanced MADDPG(AT-MADDPG)algorithm significantly improves performance,achieving notable reductions in both energy consumption and latency compared to baseline algorithms,especially in high-demand,dynamic scenarios.展开更多
Discontinuity traces significantly impact the mechanical properties of rock masses,making their rapid and accurate identification crucial for stability analysis.We propose a framework using the multi-scale surface var...Discontinuity traces significantly impact the mechanical properties of rock masses,making their rapid and accurate identification crucial for stability analysis.We propose a framework using the multi-scale surface variation index(MsSVI)and transfer-learning enhanced artificial neural network(ANN)for efficient discontinuity trace extraction from rock mass point clouds.Leveraging the similarity between regular geometric bodies and engineering rock masses,we extract trace feature points without manual threshold selection.Our contributions include:(1)An adaptive radius MsSVI calculation method based on density information;(2)a universal trace feature point classification model trained using MsSVI and ANN via inductive transfer learning;and(3)a random sampling L1-medial skeleton algorithm for precise trace feature point extraction,bypassing point cloud triangulation.Experimental results show that our model achieves a 90.2%F1-score on test sets,demonstrating its accuracy and robustness.Furthermore,our method excels in trace detail extraction on two datasets,surpassing existing models and highlighting its potential for rock mass structural analysis.展开更多
Deep neural networks(DNNs)are effective in solving both forward and inverse problems for nonlinear partial differential equations(PDEs).However,conventional DNNs are not effective in handling problems such as delay di...Deep neural networks(DNNs)are effective in solving both forward and inverse problems for nonlinear partial differential equations(PDEs).However,conventional DNNs are not effective in handling problems such as delay differential equations(DDEs)and delay integrodifferential equations(DIDEs)with constant delays,primarily due to their low regularity at delayinduced breaking points.In this paper,a DNN method that combines multi-task learning(MTL)which is proposed to solve both the forward and inverse problems of DIDEs.The core idea of this approach is to divide the original equation into multiple tasks based on the delay,using auxiliary outputs to represent the integral terms,followed by the use of MTL to seamlessly incorporate the properties at the breaking points into the loss function.Furthermore,given the increased training dificulty associated with multiple tasks and outputs,we employ a sequential training scheme to reduce training complexity and provide reference solutions for subsequent tasks.This approach significantly enhances the approximation accuracy of solving DIDEs with DNNs,as demonstrated by comparisons with traditional DNN methods.We validate the effectiveness of this method through several numerical experiments,test various parameter sharing structures in MTL and compare the testing results of these structures.Finally,this method is implemented to solve the inverse problem of nonlinear DIDE and the results show that the unknown parameters of DIDE can be discovered with sparse or noisy data.展开更多
Two-dimensional endoscopic images are susceptible to interferences such as specular reflections and monotonous texture illumination,hindering accurate three-dimensional lesion reconstruction by surgical robots.This st...Two-dimensional endoscopic images are susceptible to interferences such as specular reflections and monotonous texture illumination,hindering accurate three-dimensional lesion reconstruction by surgical robots.This study proposes a novel end-to-end disparity estimation model to address these challenges.Our approach combines a Pseudo-Siamese neural network architecture with pyramid dilated convolutions,integrating multi-scale image information to enhance robustness against lighting interferences.This study introduces a Pseudo-Siamese structure-based disparity regression model that simplifies left-right image comparison,improving accuracy and efficiency.The model was evaluated using a dataset of stereo endoscopic videos captured by the Da Vinci surgical robot,comprising simulated silicone heart sequences and real heart video data.Experimental results demonstrate significant improvement in the network’s resistance to lighting interference without substantially increasing parameters.Moreover,the model exhibited faster convergence during training,contributing to overall performance enhancement.This study advances endoscopic image processing accuracy and has potential implications for surgical robot applications in complex environments.展开更多
Adversarial Reinforcement Learning(ARL)models for intelligent devices and Network Intrusion Detection Systems(NIDS)improve systemresilience against sophisticated cyber-attacks.As a core component of ARL,Adversarial Tr...Adversarial Reinforcement Learning(ARL)models for intelligent devices and Network Intrusion Detection Systems(NIDS)improve systemresilience against sophisticated cyber-attacks.As a core component of ARL,Adversarial Training(AT)enables NIDS agents to discover and prevent newattack paths by exposing them to competing examples,thereby increasing detection accuracy,reducing False Positives(FPs),and enhancing network security.To develop robust decision-making capabilities for real-world network disruptions and hostile activity,NIDS agents are trained in adversarial scenarios to monitor the current state and notify management of any abnormal or malicious activity.The accuracy and timeliness of the IDS were crucial to the network’s availability and reliability at this time.This paper analyzes ARL applications in NIDS,revealing State-of-The-Art(SoTA)methodology,issues,and future research prospects.This includes Reinforcement Machine Learning(RML)-based NIDS,which enables an agent to interact with the environment to achieve a goal,andDeep Reinforcement Learning(DRL)-based NIDS,which can solve complex decision-making problems.Additionally,this survey study addresses cybersecurity adversarial circumstances and their importance for ARL and NIDS.Architectural design,RL algorithms,feature representation,and training methodologies are examined in the ARL-NIDS study.This comprehensive study evaluates ARL for intelligent NIDS research,benefiting cybersecurity researchers,practitioners,and policymakers.The report promotes cybersecurity defense research and innovation.展开更多
Objective To address the dual challenges of long-tail distribution and feature sparsity in traditional Chinese medicine(TCM)syndrome differentiation within real clinical settings,we propose a data-efficient learning f...Objective To address the dual challenges of long-tail distribution and feature sparsity in traditional Chinese medicine(TCM)syndrome differentiation within real clinical settings,we propose a data-efficient learning framework enhanced by knowledge graphs.Methods We developed Agent-GNN,a three-stage decoupled learning framework,and validated it on the Traditional Chinese Medicine Syndrome Diagnosis(TCM-SD)dataset containing 54152 clinical records across 148 syndrome categories.First,we constructed a comprehensive medical knowledge graph encoding the complete TCM reasoning system.Second,we proposed a Functional Patient Profiling(FPP)method that utilizes large language models(LLMs)combined with Graph Retrieval-Augmented Generation(RAG)to extract structured symptom-etiology-pathogenesis subgraphs from medical records.Third,we employed heterogeneous graph neural networks to learn structured combination patterns explicitly.We compared our method against multiple baselines including BERT,ZY-BERT,ZY-BERT+Know,GAT,and GPT-4 Few-shot,using macro-F1 score as the primary evaluation metric.Additionally,ablation experiments were conducted to validate the contribution of each key component to model performance.Results Agent-GNN achieved an overall macro-F1 score of 72.4%,representing an 8.7 percentage points improvement over ZY-BERT+Know(63.7%),the strongest baseline among traditional methods.For long-tail syndromes with fewer than 10 samples,Agent-GNN reached a macro-F1 score of 58.6%,compared with 39.3%for ZY-BERT+Know and 41.2%for GPT-4 Few-shot,representing relative improvements of 49.2%and 42.2%,respectively.Ablation experiments confirmed that the explicit modeling of etiology-pathogenesis nodes contributed 12.4 percentage points to this enhanced long-tail syndrome performance.Conclusion This study proposes Agent-GNN,a knowledge graph-enhanced framework that effectively addresses the long-tail distribution challenge in TCM syndrome differentiation.By explicitly modeling manifestation-mechanism-essence patterns through structured knowledge graphs,our approach achieves superior performance in data-scarce scenarios while providing interpretable reasoning paths for TCM intelligent diagnosis.展开更多
Mental-health risk detection seeks early signs of distress from social media posts and clinical transcripts to enable timely intervention before crises.When such risks go undetected,consequences can escalate to self-h...Mental-health risk detection seeks early signs of distress from social media posts and clinical transcripts to enable timely intervention before crises.When such risks go undetected,consequences can escalate to self-harm,long-term disability,reduced productivity,and significant societal and economic burden.Despite recent advances,detecting risk from online text remains challenging due to heterogeneous language,evolving semantics,and the sequential emergence of new datasets.Effective solutions must encode clinically meaningful cues,reason about causal relations,and adapt to new domains without forgetting prior knowledge.To address these challenges,this paper presents a Continual Neuro-Symbolic Graph Learning(CNSGL)framework that unifies symbolic reasoning,causal inference,and continual learning within a single architecture.Each post is represented as a symbolic graph linking clinically relevant tags to textual content,enriched with causal edges derived from directional Point-wise Mutual Information(PMI).A two-layer Graph Convolutional Network(GCN)encodes these graphs,and a Transformer-based attention pooler aggregates node embeddings while providing interpretable tag-level importances.Continual adaptation across datasets is achieved through the Multi-Head Freeze(MH-Freeze)strategy,which freezes a shared encoder and incrementally trains lightweight task-specific heads(small classifiers attached to the shared embedding).Experimental evaluations across six diverse mental-health datasets ranging from Reddit discourse to clinical interviews,demonstrate that MH-Freeze consistently outperforms existing continual-learning baselines in both discriminative accuracy and calibration reliability.Across six datasets,MH-Freeze achieves up to 0.925 accuracy and 0.923 F1-Score,with AUPRC≥0.934 and AUROC≥0.942,consistently surpassing all continual-learning baselines.The results confirm the framework’s ability to preserve prior knowledge,adapt to domain shifts,and maintain causal interpretability,establishing CNSGL as a promising step toward robust,explainable,and lifelong mental-health risk assessment.展开更多
Federated Learning(FL)protects data privacy through a distributed training mechanism,yet its decentralized nature also introduces new security vulnerabilities.Backdoor attacks inject malicious triggers into the global...Federated Learning(FL)protects data privacy through a distributed training mechanism,yet its decentralized nature also introduces new security vulnerabilities.Backdoor attacks inject malicious triggers into the global model through compromised updates,posing significant threats to model integrity and becoming a key focus in FL security.Existing backdoor attack methods typically embed triggers directly into original images and consider only data heterogeneity,resulting in limited stealth and adaptability.To address the heterogeneity of malicious client devices,this paper proposes a novel backdoor attack method named Capability-Adaptive Shadow Backdoor Attack(CASBA).By incorporating measurements of clients’computational and communication capabilities,CASBA employs a dynamic hierarchical attack strategy that adaptively aligns attack intensity with available resources.Furthermore,an improved deep convolutional generative adversarial network(DCGAN)is integrated into the attack pipeline to embed triggers without modifying original data,significantly enhancing stealthiness.Comparative experiments with Shadow Backdoor Attack(SBA)across multiple scenarios demonstrate that CASBA dynamically adjusts resource consumption based on device capabilities,reducing average memory usage per iteration by 5.8%.CASBA improves resource efficiency while keeping the drop in attack success rate within 3%.Additionally,the effectiveness of CASBA against three robust FL algorithms is also validated.展开更多
In Human–Robot Interaction(HRI),generating robot trajectories that accurately reflect user intentions while ensuring physical realism remains challenging,especially in unstructured environments.In this study,we devel...In Human–Robot Interaction(HRI),generating robot trajectories that accurately reflect user intentions while ensuring physical realism remains challenging,especially in unstructured environments.In this study,we develop a multimodal framework that integrates symbolic task reasoning with continuous trajectory generation.The approach employs transformer models and adversarial training to map high-level intent to robotic motion.Information from multiple data sources,such as voice traits,hand and body keypoints,visual observations,and recorded paths,is integrated simultaneously.These signals are mapped into a shared representation that supports interpretable reasoning while enabling smooth and realistic motion generation.Based on this design,two different learning strategies are investigated.In the first step,grammar-constrained Linear Temporal Logic(LTL)expressions are created from multimodal human inputs.These expressions are subsequently decoded into robot trajectories.The second method generates trajectories directly from symbolic intent and linguistic data,bypassing an intermediate logical representation.Transformer encoders combine multiple types of information,and autoregressive transformer decoders generate motion sequences.Adding smoothness and speed limits during training increases the likelihood of physical feasibility.To improve the realism and stability of the generated trajectories during training,an adversarial discriminator is also included to guide them toward the distribution of actual robot motion.Tests on the NATSGLD dataset indicate that the complete system exhibits stable training behaviour and performance.In normalised coordinates,the logic-based pipeline has an Average Displacement Error(ADE)of 0.040 and a Final Displacement Error(FDE)of 0.036.The adversarial generator makes substantially more progress,reducing ADE to 0.021 and FDE to 0.018.Visual examination confirms that the generated trajectories closely align with observed motion patterns while preserving smooth temporal dynamics.展开更多
With the increasing complexity of vehicular networks and the proliferation of connected vehicles,Federated Learning(FL)has emerged as a critical framework for decentralized model training while preserving data privacy...With the increasing complexity of vehicular networks and the proliferation of connected vehicles,Federated Learning(FL)has emerged as a critical framework for decentralized model training while preserving data privacy.However,efficient client selection and adaptive weight allocation in heterogeneous and non-IID environments remain challenging.To address these issues,we propose Federated Learning with Client Selection and Adaptive Weighting(FedCW),a novel algorithm that leverages adaptive client selection and dynamic weight allocation for optimizing model convergence in real-time vehicular networks.FedCW selects clients based on their Euclidean distance from the global model and dynamically adjusts aggregation weights to optimize both data diversity and model convergence.Experimental results show that FedCW significantly outperforms existing FL algorithms such as FedAvg,FedProx,and SCAFFOLD,particularly in non-IID settings,achieving faster convergence,higher accuracy,and reduced communication overhead.These findings demonstrate that FedCW provides an effective solution for enhancing the performance of FL in heterogeneous,edge-based computing environments.展开更多
文摘Reconfigurable intelligent surface(RIS)have been cast as a promising alternative to alleviate blockage vulnerability and enhance coverage capability for terahertz(THz)communications.Owing to large-scale array elements at transceivers and RIS,the codebook based beamforming can be utilized in a computationally efficient manner.However,the codeword selection for analog beamforming is an intractable combinatorial optimization(CO)problem.To this end,by taking the CO problem as a classification problem,a multi-task learning based analog beam selection(MTL-ABS)framework is developed to implement cooperative beam selection concurrently at transceivers and RIS.In addition,residual network and self-attention mechanism are used to combat the network degradation and mine intrinsic THz channel features.Finally,the network convergence is analyzed from a blockwise perspective,and numerical results demonstrate that the MTL-ABS framework greatly decreases the beam selection overhead and achieves near optimal sum-rate compared with heuristic search based counterparts.
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.
基金supported by the National Natural Science Foundation of China(Grant Nos.42130719 and 42177173)the Doctoral Direct Train Project of Chongqing Natural Science Foundation(Grant No.CSTB2023NSCQ-BSX0029).
文摘Underground engineering projects such as deep tunnel excavation often encounter rockburst disasters accompanied by numerous microseismic events.Rapid interpretation of microseismic signals is crucial for the timely identification of rockbursts.However,conventional processing encompasses multi-step workflows,including classification,denoising,picking,locating,and computational analysis,coupled with manual intervention,which collectively compromise the reliability of early warnings.To address these challenges,this study innovatively proposes the“microseismic stethoscope"-a multi-task machine learning and deep learning model designed for the automated processing of massive microseismic signals.This model efficiently extracts three key parameters that are necessary for recognizing rockburst disasters:rupture location,microseismic energy,and moment magnitude.Specifically,the model extracts raw waveform features from three dedicated sub-networks:a classifier for source zone classification,and two regressors for microseismic energy and moment magnitude estimation.This model demonstrates superior efficiency compared to traditional processing and semi-automated processing,reducing per-event processing time from 0.71 s to 0.49 s to merely 0.036 s.It concurrently achieves 98%accuracy in source zone classification,with microseismic energy and moment magnitude estimation errors of 0.13 and 0.05,respectively.This model has been well applied and validated in the Daxiagu Tunnel case in Sichuan,China.The application results indicate that the model is as accurate as traditional methods in determining source parameters,and thus can be used to identify potential geomechanical processes of rockburst disasters.By enhancing the signal processing reliability of microseismic events,the proposed model in this study presents a significant advancement in the identification of rockburst disasters.
文摘With the proliferation of Internet of Things(IoT)devices,securing these interconnected systems against cyberattacks has become a critical challenge.Traditional security paradigms often fail to cope with the scale and diversity of IoT network traffic.This paper presents a comparative benchmark of classic machine learning(ML)and state-of-the-art deep learning(DL)algorithms for IoT intrusion detection.Our methodology employs a twophased approach:a preliminary pilot study using a custom-generated dataset to establish baselines,followed by a comprehensive evaluation on the large-scale CICIoTDataset2023.We benchmarked algorithms including Random Forest,XGBoost,CNN,and StackedLSTM.The results indicate that while top-performingmodels frombothcategories achieve over 99%classification accuracy,this metric masks a crucial performance trade-off.We demonstrate that treebased ML ensembles exhibit superior precision(91%)in identifying benign traffic,making them effective at reducing false positives.Conversely,DL models demonstrate superior recall(96%),making them better suited for minimizing the interruption of legitimate traffic.We conclude that the selection of an optimal model is not merely a matter of maximizing accuracy but is a strategic choice dependent on the specific security priority either minimizing false alarms or ensuring service availability.Thiswork provides a practical framework for deploying context-aware security solutions in diverse IoT environments.
基金supported by the Research year project of the KongjuNational University in 2025 and the Institute of Information&Communications Technology Planning&Evaluation(IITP)grant funded by the Korea government(MSIT)(No.RS-2024-00444170,Research and International Collaboration on Trust Model-Based Intelligent Incident Response Technologies in 6G Open Network Environment).
文摘With the growing complexity and decentralization of network systems,the attack surface has expanded,which has led to greater concerns over network threats.In this context,artificial intelligence(AI)-based network intrusion detection systems(NIDS)have been extensively studied,and recent efforts have shifted toward integrating distributed learning to enable intelligent and scalable detection mechanisms.However,most existing works focus on individual distributed learning frameworks,and there is a lack of systematic evaluations that compare different algorithms under consistent conditions.In this paper,we present a comprehensive evaluation of representative distributed learning frameworks—Federated Learning(FL),Split Learning(SL),hybrid collaborative learning(SFL),and fully distributed learning—in the context of AI-driven NIDS.Using recent benchmark intrusion detection datasets,a unified model backbone,and controlled distributed scenarios,we assess these frameworks across multiple criteria,including detection performance,communication cost,computational efficiency,and convergence behavior.Our findings highlight distinct trade-offs among the distributed learning frameworks,demonstrating that the optimal choice depends strongly on systemconstraints such as bandwidth availability,node resources,and data distribution.This work provides the first holistic analysis of distributed learning approaches for AI-driven NIDS and offers practical guidelines for designing secure and efficient intrusion detection systems in decentralized environments.
基金supported by the Ministry of Education(MOE)Singapore,Academic Research Fund(AcRF)Tier 1(RG65/22)。
文摘Convolutional neural networks(CNNs)have shown remarkable success across numerous tasks such as image classification,yet the theoretical understanding of their convergence remains underdeveloped compared to their empirical achievements.In this paper,the first filter learning framework with convergence-guaranteed learning laws for end-to-end learning of deep CNNs is proposed.Novel update laws with convergence analysis are formulated based on the mathematical representation of each layer in convolutional neural networks.The proposed learning laws enable concurrent updates of weights across all layers of the deep convolutional neural network and the analysis shows that the training errors converge to certain bounds which are dependent on the approximation errors.Case studies are conducted on benchmark datasets and the results show that the proposed concurrent filter learning framework guarantees the convergence and offers more consistent and reliable results during training with a trade-off in performance compared to stochastic gradient descent methods.This framework represents a significant step towards enhancing the reliability and effectiveness of deep convolutional neural network by developing a theoretical analysis which allows practical implementation of the learning laws with automatic tuning of the learning rate to guarantee the convergence during training.
基金supported by the Extral High Voltage Power Transmission Company,China Southern Power Grid Co.,Ltd.
文摘Distributed Denial-of-Service(DDoS)attacks pose severe threats to Industrial Control Networks(ICNs),where service disruption can cause significant economic losses and operational risks.Existing signature-based methods are ineffective against novel attacks,and traditional machine learning models struggle to capture the complex temporal dependencies and dynamic traffic patterns inherent in ICN environments.To address these challenges,this study proposes a deep feature-driven hybrid framework that integrates Transformer,BiLSTM,and KNN to achieve accurate and robust DDoS detection.The Transformer component extracts global temporal dependencies from network traffic flows,while BiLSTM captures fine-grained sequential dynamics.The learned embeddings are then classified using an instance-based KNN layer,enhancing decision boundary precision.This cascaded architecture balances feature abstraction and locality preservation,improving both generalization and robustness.The proposed approach was evaluated on a newly collected real-time ICN traffic dataset and further validated using the public CIC-IDS2017 and Edge-IIoT datasets to demonstrate generalization.Comprehensive metrics including accuracy,precision,recall,F1-score,ROC-AUC,PR-AUC,false positive rate(FPR),and detection latency were employed.Results show that the hybrid framework achieves 98.42%accuracy with an ROC-AUC of 0.992 and FPR below 1%,outperforming baseline machine learning and deep learning models.Robustness experiments under Gaussian noise perturbations confirmed stable performance with less than 2%accuracy degradation.Moreover,detection latency remained below 2.1 ms per sample,indicating suitability for real-time ICS deployment.In summary,the proposed hybrid temporal learning and instance-based classification model offers a scalable and effective solution for DDoS detection in industrial control environments.By combining global contextual modeling,sequential learning,and instance-based refinement,the framework demonstrates strong adaptability across datasets and resilience against noise,providing practical utility for safeguarding critical infrastructure.
文摘As the types of traffic requests increase,the elastic optical network(EON)is considered as a promising architecture to carry multiple types of traffic requests simultaneously,including immediate reservation(IR)and advance reservation(AR).Various resource allocation schemes for IR/AR requests have been designed in EON to reduce bandwidth blocking probability(BBP).However,these schemes do not consider different transmission requirements of IR requests and cannot maintain a low BBP for high-priority requests.In this paper,multi-priority is considered in the hybrid IR/AR request scenario.We modify the asynchronous advantage actor critic(A3C)model and propose an A3C-assisted priority resource allocation(APRA)algorithm.The APRA integrates priority and transmission quality of IR requests to design the A3C reward function,then dynamically allocates dedicated resources for different IR requests according to the time-varying requirements.By maximizing the reward,the transmission quality of IR requests can be matched with the priority,and lower BBP for high-priority IR requests can be ensured.Simulation results show that the APRA reduces the BBP of high-priority IR requests from 0.0341 to0.0138,and the overall network operation gain is improved by 883 compared to the scheme without considering the priority.
基金funded by the Deanship of Scientific Research(DSR)at King Abdulaziz University,Jeddah,Saudi Arabia,grant number RG-2-611-42(A.O.A.).
文摘Wi-Fi technology has evolved significantly since its introduction in 1997,advancing to Wi-Fi 6 as the latest standard,with Wi-Fi 7 currently under development.Despite these advancements,integrating machine learning into Wi-Fi networks remains challenging,especially in decentralized environments with multiple access points(mAPs).This paper is a short review that summarizes the potential applications of federated reinforcement learning(FRL)across eight key areas of Wi-Fi functionality,including channel access,link adaptation,beamforming,multi-user transmissions,channel bonding,multi-link operation,spatial reuse,and multi-basic servic set(multi-BSS)coordination.FRL is highlighted as a promising framework for enabling decentralized training and decision-making while preserving data privacy.To illustrate its role in practice,we present a case study on link activation in a multi-link operation(MLO)environment with multiple APs.Through theoretical discussion and simulation results,the study demonstrates how FRL can improve performance and reliability,paving the way for more adaptive and collaborative Wi-Fi networks in the era of Wi-Fi 7 and beyond.
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supported by the National Key Research and Development Program of China under grant 2021YFA0716600。
文摘In response to the rising demand for low-latency,computation-intensive applications in vehicular networks,this paper proposes an adaptive task offloading approach for Vehicle-to-Everything(V2X)environments.Leveraging an enhanced Multi-Agent Deep Deterministic Policy Gradient(MADDPG)algorithm with an attention mechanism,the proposed approach optimizes computation offloading and resource allocation,aiming to minimize energy consumption and service delay.In this paper,vehicles dynamically offload computing-intensive tasks to both nearby vehicles through V2V links and roadside units through V2I links.The adaptive attention mechanism enables the system to prioritize relevant state information,leading to faster convergence.Simulations conducted in a realistic urban V2X scenario demonstrate that the proposed Attention-enhanced MADDPG(AT-MADDPG)algorithm significantly improves performance,achieving notable reductions in both energy consumption and latency compared to baseline algorithms,especially in high-demand,dynamic scenarios.
基金supported by the Fundamental Research Funds for the Central Universities(Grant No.2025XJSB01)the Founda-tion of State Key Laboratory for Geomechanics and Deep Under-ground Engineering,China University of Mining&Technology,Beijing.(Grant No.SKLGDUEK 2217)the Collaborative Inno-vation Center for Prevention and Control of Mountain Geological Hazards of Zhejiang Province(PCMGH-2022-03).
文摘Discontinuity traces significantly impact the mechanical properties of rock masses,making their rapid and accurate identification crucial for stability analysis.We propose a framework using the multi-scale surface variation index(MsSVI)and transfer-learning enhanced artificial neural network(ANN)for efficient discontinuity trace extraction from rock mass point clouds.Leveraging the similarity between regular geometric bodies and engineering rock masses,we extract trace feature points without manual threshold selection.Our contributions include:(1)An adaptive radius MsSVI calculation method based on density information;(2)a universal trace feature point classification model trained using MsSVI and ANN via inductive transfer learning;and(3)a random sampling L1-medial skeleton algorithm for precise trace feature point extraction,bypassing point cloud triangulation.Experimental results show that our model achieves a 90.2%F1-score on test sets,demonstrating its accuracy and robustness.Furthermore,our method excels in trace detail extraction on two datasets,surpassing existing models and highlighting its potential for rock mass structural analysis.
文摘Deep neural networks(DNNs)are effective in solving both forward and inverse problems for nonlinear partial differential equations(PDEs).However,conventional DNNs are not effective in handling problems such as delay differential equations(DDEs)and delay integrodifferential equations(DIDEs)with constant delays,primarily due to their low regularity at delayinduced breaking points.In this paper,a DNN method that combines multi-task learning(MTL)which is proposed to solve both the forward and inverse problems of DIDEs.The core idea of this approach is to divide the original equation into multiple tasks based on the delay,using auxiliary outputs to represent the integral terms,followed by the use of MTL to seamlessly incorporate the properties at the breaking points into the loss function.Furthermore,given the increased training dificulty associated with multiple tasks and outputs,we employ a sequential training scheme to reduce training complexity and provide reference solutions for subsequent tasks.This approach significantly enhances the approximation accuracy of solving DIDEs with DNNs,as demonstrated by comparisons with traditional DNN methods.We validate the effectiveness of this method through several numerical experiments,test various parameter sharing structures in MTL and compare the testing results of these structures.Finally,this method is implemented to solve the inverse problem of nonlinear DIDE and the results show that the unknown parameters of DIDE can be discovered with sparse or noisy data.
基金Supported by Sichuan Science and Technology Program(2023YFSY0026,2023YFH0004)Supported by the Institute of Information&Communications Technology Planning&Evaluation(IITP)grant funded by the Korean government(MSIT)(No.RS-2022-00155885,Artificial Intelligence Convergence Innovation Human Resources Development(Hanyang University ERICA)).
文摘Two-dimensional endoscopic images are susceptible to interferences such as specular reflections and monotonous texture illumination,hindering accurate three-dimensional lesion reconstruction by surgical robots.This study proposes a novel end-to-end disparity estimation model to address these challenges.Our approach combines a Pseudo-Siamese neural network architecture with pyramid dilated convolutions,integrating multi-scale image information to enhance robustness against lighting interferences.This study introduces a Pseudo-Siamese structure-based disparity regression model that simplifies left-right image comparison,improving accuracy and efficiency.The model was evaluated using a dataset of stereo endoscopic videos captured by the Da Vinci surgical robot,comprising simulated silicone heart sequences and real heart video data.Experimental results demonstrate significant improvement in the network’s resistance to lighting interference without substantially increasing parameters.Moreover,the model exhibited faster convergence during training,contributing to overall performance enhancement.This study advances endoscopic image processing accuracy and has potential implications for surgical robot applications in complex environments.
文摘Adversarial Reinforcement Learning(ARL)models for intelligent devices and Network Intrusion Detection Systems(NIDS)improve systemresilience against sophisticated cyber-attacks.As a core component of ARL,Adversarial Training(AT)enables NIDS agents to discover and prevent newattack paths by exposing them to competing examples,thereby increasing detection accuracy,reducing False Positives(FPs),and enhancing network security.To develop robust decision-making capabilities for real-world network disruptions and hostile activity,NIDS agents are trained in adversarial scenarios to monitor the current state and notify management of any abnormal or malicious activity.The accuracy and timeliness of the IDS were crucial to the network’s availability and reliability at this time.This paper analyzes ARL applications in NIDS,revealing State-of-The-Art(SoTA)methodology,issues,and future research prospects.This includes Reinforcement Machine Learning(RML)-based NIDS,which enables an agent to interact with the environment to achieve a goal,andDeep Reinforcement Learning(DRL)-based NIDS,which can solve complex decision-making problems.Additionally,this survey study addresses cybersecurity adversarial circumstances and their importance for ARL and NIDS.Architectural design,RL algorithms,feature representation,and training methodologies are examined in the ARL-NIDS study.This comprehensive study evaluates ARL for intelligent NIDS research,benefiting cybersecurity researchers,practitioners,and policymakers.The report promotes cybersecurity defense research and innovation.
基金Sichuan TCM Culture Coordinated Development Research Center Project(2023XT131)National Key Science and Technology Project of China(2023ZD0509405)National Natural Science Foundation of China(82174236).
文摘Objective To address the dual challenges of long-tail distribution and feature sparsity in traditional Chinese medicine(TCM)syndrome differentiation within real clinical settings,we propose a data-efficient learning framework enhanced by knowledge graphs.Methods We developed Agent-GNN,a three-stage decoupled learning framework,and validated it on the Traditional Chinese Medicine Syndrome Diagnosis(TCM-SD)dataset containing 54152 clinical records across 148 syndrome categories.First,we constructed a comprehensive medical knowledge graph encoding the complete TCM reasoning system.Second,we proposed a Functional Patient Profiling(FPP)method that utilizes large language models(LLMs)combined with Graph Retrieval-Augmented Generation(RAG)to extract structured symptom-etiology-pathogenesis subgraphs from medical records.Third,we employed heterogeneous graph neural networks to learn structured combination patterns explicitly.We compared our method against multiple baselines including BERT,ZY-BERT,ZY-BERT+Know,GAT,and GPT-4 Few-shot,using macro-F1 score as the primary evaluation metric.Additionally,ablation experiments were conducted to validate the contribution of each key component to model performance.Results Agent-GNN achieved an overall macro-F1 score of 72.4%,representing an 8.7 percentage points improvement over ZY-BERT+Know(63.7%),the strongest baseline among traditional methods.For long-tail syndromes with fewer than 10 samples,Agent-GNN reached a macro-F1 score of 58.6%,compared with 39.3%for ZY-BERT+Know and 41.2%for GPT-4 Few-shot,representing relative improvements of 49.2%and 42.2%,respectively.Ablation experiments confirmed that the explicit modeling of etiology-pathogenesis nodes contributed 12.4 percentage points to this enhanced long-tail syndrome performance.Conclusion This study proposes Agent-GNN,a knowledge graph-enhanced framework that effectively addresses the long-tail distribution challenge in TCM syndrome differentiation.By explicitly modeling manifestation-mechanism-essence patterns through structured knowledge graphs,our approach achieves superior performance in data-scarce scenarios while providing interpretable reasoning paths for TCM intelligent diagnosis.
基金supported by the National Research Foundation of Korea(NRF)grant funded by the Korea government(MSIT)(RS-2025-00518960)in part by the National Research Foundation of Korea(NRF)grant funded by the Korea government(MSIT)(RS-2025-00563192).
文摘Mental-health risk detection seeks early signs of distress from social media posts and clinical transcripts to enable timely intervention before crises.When such risks go undetected,consequences can escalate to self-harm,long-term disability,reduced productivity,and significant societal and economic burden.Despite recent advances,detecting risk from online text remains challenging due to heterogeneous language,evolving semantics,and the sequential emergence of new datasets.Effective solutions must encode clinically meaningful cues,reason about causal relations,and adapt to new domains without forgetting prior knowledge.To address these challenges,this paper presents a Continual Neuro-Symbolic Graph Learning(CNSGL)framework that unifies symbolic reasoning,causal inference,and continual learning within a single architecture.Each post is represented as a symbolic graph linking clinically relevant tags to textual content,enriched with causal edges derived from directional Point-wise Mutual Information(PMI).A two-layer Graph Convolutional Network(GCN)encodes these graphs,and a Transformer-based attention pooler aggregates node embeddings while providing interpretable tag-level importances.Continual adaptation across datasets is achieved through the Multi-Head Freeze(MH-Freeze)strategy,which freezes a shared encoder and incrementally trains lightweight task-specific heads(small classifiers attached to the shared embedding).Experimental evaluations across six diverse mental-health datasets ranging from Reddit discourse to clinical interviews,demonstrate that MH-Freeze consistently outperforms existing continual-learning baselines in both discriminative accuracy and calibration reliability.Across six datasets,MH-Freeze achieves up to 0.925 accuracy and 0.923 F1-Score,with AUPRC≥0.934 and AUROC≥0.942,consistently surpassing all continual-learning baselines.The results confirm the framework’s ability to preserve prior knowledge,adapt to domain shifts,and maintain causal interpretability,establishing CNSGL as a promising step toward robust,explainable,and lifelong mental-health risk assessment.
基金supported by the National Natural Science Foundation of China(Grant No.62172123)the Key Research and Development Program of Heilongjiang Province,China(GrantNo.2022ZX01A36).
文摘Federated Learning(FL)protects data privacy through a distributed training mechanism,yet its decentralized nature also introduces new security vulnerabilities.Backdoor attacks inject malicious triggers into the global model through compromised updates,posing significant threats to model integrity and becoming a key focus in FL security.Existing backdoor attack methods typically embed triggers directly into original images and consider only data heterogeneity,resulting in limited stealth and adaptability.To address the heterogeneity of malicious client devices,this paper proposes a novel backdoor attack method named Capability-Adaptive Shadow Backdoor Attack(CASBA).By incorporating measurements of clients’computational and communication capabilities,CASBA employs a dynamic hierarchical attack strategy that adaptively aligns attack intensity with available resources.Furthermore,an improved deep convolutional generative adversarial network(DCGAN)is integrated into the attack pipeline to embed triggers without modifying original data,significantly enhancing stealthiness.Comparative experiments with Shadow Backdoor Attack(SBA)across multiple scenarios demonstrate that CASBA dynamically adjusts resource consumption based on device capabilities,reducing average memory usage per iteration by 5.8%.CASBA improves resource efficiency while keeping the drop in attack success rate within 3%.Additionally,the effectiveness of CASBA against three robust FL algorithms is also validated.
基金The authors extend their appreciation to Prince Sattam bin Abdulaziz University for funding this research work through the project number(PSAU/2024/01/32082).
文摘In Human–Robot Interaction(HRI),generating robot trajectories that accurately reflect user intentions while ensuring physical realism remains challenging,especially in unstructured environments.In this study,we develop a multimodal framework that integrates symbolic task reasoning with continuous trajectory generation.The approach employs transformer models and adversarial training to map high-level intent to robotic motion.Information from multiple data sources,such as voice traits,hand and body keypoints,visual observations,and recorded paths,is integrated simultaneously.These signals are mapped into a shared representation that supports interpretable reasoning while enabling smooth and realistic motion generation.Based on this design,two different learning strategies are investigated.In the first step,grammar-constrained Linear Temporal Logic(LTL)expressions are created from multimodal human inputs.These expressions are subsequently decoded into robot trajectories.The second method generates trajectories directly from symbolic intent and linguistic data,bypassing an intermediate logical representation.Transformer encoders combine multiple types of information,and autoregressive transformer decoders generate motion sequences.Adding smoothness and speed limits during training increases the likelihood of physical feasibility.To improve the realism and stability of the generated trajectories during training,an adversarial discriminator is also included to guide them toward the distribution of actual robot motion.Tests on the NATSGLD dataset indicate that the complete system exhibits stable training behaviour and performance.In normalised coordinates,the logic-based pipeline has an Average Displacement Error(ADE)of 0.040 and a Final Displacement Error(FDE)of 0.036.The adversarial generator makes substantially more progress,reducing ADE to 0.021 and FDE to 0.018.Visual examination confirms that the generated trajectories closely align with observed motion patterns while preserving smooth temporal dynamics.
文摘With the increasing complexity of vehicular networks and the proliferation of connected vehicles,Federated Learning(FL)has emerged as a critical framework for decentralized model training while preserving data privacy.However,efficient client selection and adaptive weight allocation in heterogeneous and non-IID environments remain challenging.To address these issues,we propose Federated Learning with Client Selection and Adaptive Weighting(FedCW),a novel algorithm that leverages adaptive client selection and dynamic weight allocation for optimizing model convergence in real-time vehicular networks.FedCW selects clients based on their Euclidean distance from the global model and dynamically adjusts aggregation weights to optimize both data diversity and model convergence.Experimental results show that FedCW significantly outperforms existing FL algorithms such as FedAvg,FedProx,and SCAFFOLD,particularly in non-IID settings,achieving faster convergence,higher accuracy,and reduced communication overhead.These findings demonstrate that FedCW provides an effective solution for enhancing the performance of FL in heterogeneous,edge-based computing environments.
