A Dynamic Vertical Scaling Strategy for Distributed Stream Processing Applications in Edge Computing
本文针对边缘计算中分布式流处理应用的资源限制问题,提出了一种基于近端策略优化的动态垂直扩展策略,以保证低延迟和高吞吐量。
本文针对边缘计算中分布式流处理应用的资源限制问题,提出了一种基于近端策略优化的动态垂直扩展策略,以保证低延迟和高吞吐量。
该研究使用CVAEs方法生成具有不同强度的真实感面部表情,解决了虚拟人物表情生成的挑战,且在少量数据下保持了情感表达的一致性和可控性。
Online misinformation detection faces challenges in scalability and reliance on external knowledge. This work proposes a novel method that requires no fine-tuning, retrieval, or task-specific supervision, treating truthfulness as a geometric property within the representation space of pretrained language models. By contrasting activation patterns between true and false statements, the approach identifies a “falsehood direction” in the residual stream and classifies inputs via projection of their final-layer activations. Combining contrastive activation addition (CAA) with an MLP classifier, the method demonstrates strong performance across mainstream architectures—including Gemma, Llama, and Qwen—matching or surpassing zero- and few-shot prompting on benchmarks LIAR and FACTors, with particularly notable gains for smaller models. Its performance is limited on AVeriTeC, which relies on annotated evidence, underscoring the method’s paradigm of evidence-free detection.
Traditional statistical features often fail to adequately capture the complex and dynamic nature of network traffic, thereby limiting the performance of anomaly detection systems. To address this limitation, this work proposes incorporating entropy as a lightweight and interpretable supplementary feature to quantify the variability of traffic attributes. The entropy-based feature is combined with conventional statistical features and fed into standard machine learning classifiers without replacing existing feature engineering pipelines. Experimental results on publicly available intrusion detection datasets demonstrate that this approach significantly improves classification accuracy, particularly in high-variability scenarios where it effectively reduces false positive rates. Moreover, the method incurs minimal computational overhead, offering a practical and scalable solution for real-world deployment.
Urban expansion has led to habitat fragmentation for arboreal species, while conventional manual review of camera trap imagery remains inefficient and prone to false positives. To address this, this study presents the first application of YOLOv10 for automated detection of brown howler monkeys, fine-tuning the model using a combination of camera trap videos and auxiliary image data. The approach systematically evaluates the added value of auxiliary data in few-shot wildlife identification scenarios. Results demonstrate that the proposed method significantly improves detection accuracy and computational efficiency, substantially reducing the burden of manual annotation. Furthermore, it enables effective automated monitoring of canopy corridor use, thereby providing a robust technical foundation for evaluating the efficacy of conservation interventions.
本文针对边缘计算中分布式流处理应用的资源限制问题,提出了一种基于近端策略优化的动态垂直扩展策略,以保证低延迟和高吞吐量。
该研究使用CVAEs方法生成具有不同强度的真实感面部表情,解决了虚拟人物表情生成的挑战,且在少量数据下保持了情感表达的一致性和可控性。
Online misinformation detection faces challenges in scalability and reliance on external knowledge. This work proposes a novel method that requires no fine-tuning, retrieval, or task-specific supervision, treating truthfulness as a geometric property within the representation space of pretrained language models. By contrasting activation patterns between true and false statements, the approach identifies a “falsehood direction” in the residual stream and classifies inputs via projection of their final-layer activations. Combining contrastive activation addition (CAA) with an MLP classifier, the method demonstrates strong performance across mainstream architectures—including Gemma, Llama, and Qwen—matching or surpassing zero- and few-shot prompting on benchmarks LIAR and FACTors, with particularly notable gains for smaller models. Its performance is limited on AVeriTeC, which relies on annotated evidence, underscoring the method’s paradigm of evidence-free detection.
Traditional statistical features often fail to adequately capture the complex and dynamic nature of network traffic, thereby limiting the performance of anomaly detection systems. To address this limitation, this work proposes incorporating entropy as a lightweight and interpretable supplementary feature to quantify the variability of traffic attributes. The entropy-based feature is combined with conventional statistical features and fed into standard machine learning classifiers without replacing existing feature engineering pipelines. Experimental results on publicly available intrusion detection datasets demonstrate that this approach significantly improves classification accuracy, particularly in high-variability scenarios where it effectively reduces false positive rates. Moreover, the method incurs minimal computational overhead, offering a practical and scalable solution for real-world deployment.
Urban expansion has led to habitat fragmentation for arboreal species, while conventional manual review of camera trap imagery remains inefficient and prone to false positives. To address this, this study presents the first application of YOLOv10 for automated detection of brown howler monkeys, fine-tuning the model using a combination of camera trap videos and auxiliary image data. The approach systematically evaluates the added value of auxiliary data in few-shot wildlife identification scenarios. Results demonstrate that the proposed method significantly improves detection accuracy and computational efficiency, substantially reducing the burden of manual annotation. Furthermore, it enables effective automated monitoring of canopy corridor use, thereby providing a robust technical foundation for evaluating the efficacy of conservation interventions.