CED-EF: Compressed Exact Diffusion with Error Feedback for Multi-Agent Learning
本文提出CED-EF方法，通过误差反馈和压缩通信解决多智能体学习中的去中心化随机优化问题，改善了对压缩级别和网络条件的依赖。
0 citationsRead paper
Selected work
Representative Papers
GeoRisk-RAG: A Hierarchy-Aware Risk Framework for Improving RAG Reliability through Selective Answering
本文提出GeoRisk-RAG框架，通过选择性回答和DAG距离估计地理适用性，提高基于RAG的模型在地理相关问题上的可靠性和准确性。
0 citationsRead paper
Stochastic Gradient Tracking over Time-Varying Networks: One-Step Lyapunov Analysis
This study addresses the analytical challenges in establishing convergence for decentralized stochastic gradient tracking over time-varying networks by proposing a single-step Lyapunov analysis framework. By constructing a time-varying quadratic norm and deriving a single-step Lyapunov identity, this approach circumvents traditional window expansion techniques under window mixing conditions, enabling precise recursive characterization of coupled errors. The proposed method effectively resolves critical analytical bottlenecks associated with dynamic topologies. Theoretical results demonstrate that the algorithm achieves a convergence rate matching centralized mini-batch methods, thereby attaining linear speedup. Consequently, this work provides a more concise and rigorous theoretical foundation for distributed optimization in time-varying network environments.
0 citationsRead paper
Evaluation of Adversarial Robustness in Arabic Language Models
This study systematically evaluates the security and robustness of state-of-the-art Arabic language models under multi-level adversarial attacks. Addressing the morphological complexity unique to Arabic, the work introduces the first multi-granularity adversarial benchmark encompassing character-level (e.g., diacritic insertion), word-level (e.g., conjunction manipulation), and sentence-level (e.g., semantics-preserving paraphrasing) attacks, and enhances model defenses through adversarial training. Experimental results reveal that diacritic-based attacks can degrade model accuracy by up to 92%. Adversarial training substantially improves robustness, with MARBERT demonstrating the strongest overall resilience and AraBERT achieving the largest relative gains. Nevertheless, character-level noise remains a critical vulnerability. This research uncovers distinct adversarial fragility patterns inherent to morphologically rich languages and offers new insights for securing Arabic NLP systems.
0 citationsRead paper
Cooperative RSU Sleep Scheduling for Green V2I Corridors
This study addresses the energy inefficiency of roadside units (RSUs) in vehicle-to-infrastructure (V2I) networks caused by continuous operation during off-peak hours, as well as the risk of excessive communication latency due to wake-up delays under independent sleep scheduling. To overcome these challenges, the authors propose a spatially correlated cooperative sleep scheduling framework that leverages upstream traffic detection signals shared across infrastructure links and exploits spatiotemporal correlations in vehicular traffic at adjacent intersections to enable predictive wake-up. The problem is formulated as a constrained Markov decision process and efficiently solved by decomposition into single-RSU subproblems. Evaluated on real-world traffic data from four intersections in Kuwait City, the approach achieves a 59.5% reduction in energy consumption while maintaining a 99% latency compliance rate. Extrapolated to a 200-RSU deployment scenario, it yields an annual carbon reduction of 5.25 metric tons, representing a 7.7% additional energy saving over independent optimization.
0 citationsRead paper
Recent publications
Latest Papers
CED-EF: Compressed Exact Diffusion with Error Feedback for Multi-Agent Learning
本文提出CED-EF方法，通过误差反馈和压缩通信解决多智能体学习中的去中心化随机优化问题，改善了对压缩级别和网络条件的依赖。
0 citationsRead paper
GeoRisk-RAG: A Hierarchy-Aware Risk Framework for Improving RAG Reliability through Selective Answering
本文提出GeoRisk-RAG框架，通过选择性回答和DAG距离估计地理适用性，提高基于RAG的模型在地理相关问题上的可靠性和准确性。
0 citationsRead paper
Stochastic Gradient Tracking over Time-Varying Networks: One-Step Lyapunov Analysis
This study addresses the analytical challenges in establishing convergence for decentralized stochastic gradient tracking over time-varying networks by proposing a single-step Lyapunov analysis framework. By constructing a time-varying quadratic norm and deriving a single-step Lyapunov identity, this approach circumvents traditional window expansion techniques under window mixing conditions, enabling precise recursive characterization of coupled errors. The proposed method effectively resolves critical analytical bottlenecks associated with dynamic topologies. Theoretical results demonstrate that the algorithm achieves a convergence rate matching centralized mini-batch methods, thereby attaining linear speedup. Consequently, this work provides a more concise and rigorous theoretical foundation for distributed optimization in time-varying network environments.
0 citationsRead paper
Evaluation of Adversarial Robustness in Arabic Language Models
This study systematically evaluates the security and robustness of state-of-the-art Arabic language models under multi-level adversarial attacks. Addressing the morphological complexity unique to Arabic, the work introduces the first multi-granularity adversarial benchmark encompassing character-level (e.g., diacritic insertion), word-level (e.g., conjunction manipulation), and sentence-level (e.g., semantics-preserving paraphrasing) attacks, and enhances model defenses through adversarial training. Experimental results reveal that diacritic-based attacks can degrade model accuracy by up to 92%. Adversarial training substantially improves robustness, with MARBERT demonstrating the strongest overall resilience and AraBERT achieving the largest relative gains. Nevertheless, character-level noise remains a critical vulnerability. This research uncovers distinct adversarial fragility patterns inherent to morphologically rich languages and offers new insights for securing Arabic NLP systems.
0 citationsRead paper
Cooperative RSU Sleep Scheduling for Green V2I Corridors
This study addresses the energy inefficiency of roadside units (RSUs) in vehicle-to-infrastructure (V2I) networks caused by continuous operation during off-peak hours, as well as the risk of excessive communication latency due to wake-up delays under independent sleep scheduling. To overcome these challenges, the authors propose a spatially correlated cooperative sleep scheduling framework that leverages upstream traffic detection signals shared across infrastructure links and exploits spatiotemporal correlations in vehicular traffic at adjacent intersections to enable predictive wake-up. The problem is formulated as a constrained Markov decision process and efficiently solved by decomposition into single-RSU subproblems. Evaluated on real-world traffic data from four intersections in Kuwait City, the approach achieves a 59.5% reduction in energy consumption while maintaining a 99% latency compliance rate. Extrapolated to a 200-RSU deployment scenario, it yields an annual carbon reduction of 5.25 metric tons, representing a 7.7% additional energy saving over independent optimization.
0 citationsRead paper