现有基于预训练语言模型(PLM)的作文自动评分(AES)方法偏向于直接使用从PLM提取的全局语义特征表示作文的质量,却忽略了作文质量与更细粒度特征关联关系的问题。聚焦于中文AES研究,从多种文本角度分析和评估作文质量,提出利用图神经网络...现有基于预训练语言模型(PLM)的作文自动评分(AES)方法偏向于直接使用从PLM提取的全局语义特征表示作文的质量,却忽略了作文质量与更细粒度特征关联关系的问题。聚焦于中文AES研究,从多种文本角度分析和评估作文质量,提出利用图神经网络(GNN)对作文的多尺度特征进行联合学习的中文AES方法。首先,利用GNN分别获取作文在句子级别和段落级别的篇章特征;然后,将这些篇章特征与作文的全局语义特征进行联合特征学习,实现对作文更精准的评分;最后,构建一个中文AES数据集,为中文AES研究提供数据基础。在所构建的数据集上的实验结果表明,所提方法在6个作文主题上的平均二次加权Kappa(QWK)系数相较于R2-BERT(Bidirectional Encoder Representations from Transformers model with Regression and Ranking)提升了1.1个百分点,验证了在AES任务中进行多尺度特征联合学习的有效性。同时,消融实验结果进一步表明了不同尺度的作文特征对评分效果的贡献。为了证明小模型在特定任务场景下的优越性,与当前流行的通用大语言模型GPT-3.5-turbo和DeepSeek-V3进行了对比。结果表明,使用所提方法的BERT(Bidirectional Encoder Representations from Transformers)模型在6个作文主题上的平均QWK比GPT-3.5-turbo和DeepSeek-V3分别高出了65.8和45.3个百分点,验证了大语言模型(LLMs)在面向领域的篇章级作文评分任务中,因缺乏大规模有监督微调数据而表现不佳的观点。展开更多
利用光学和SAR遥感影像进行建筑高度估计对于理解城市形态和优化城市存量空间具有重要意义。然而,现有的数据集存在诸多局限:由于样本数量较少,难以满足基于深度学习的遥感信息提取需求,样本所覆盖的区域较为有限,无法提供足够的地理多...利用光学和SAR遥感影像进行建筑高度估计对于理解城市形态和优化城市存量空间具有重要意义。然而,现有的数据集存在诸多局限:由于样本数量较少,难以满足基于深度学习的遥感信息提取需求,样本所覆盖的区域较为有限,无法提供足够的地理多样性和空间特征代表性,特别是针对中国区域的大规模建筑高度数据集尤为缺乏。此外,数据集的开源性不足,限制了其在更广泛的研究中的应用和验证。为解决这些问题,本文构建了一个面向深度学习的基于Sentinel影像的建筑物高度数据集BHDSI(Building Height Estimation Dataset Based on Sentinel Imagery),该数据集涵盖了中国62个城市的中心城区,共有5606个样本,覆盖了城市,农村等场景,是目前中国区域覆盖面积最大的建筑高度数据集。该数据集包含哨兵一号和哨兵二号的遥感影像以及建筑高度的真实值,样本大小是256×256,相比于64×64大小的数据集,为建筑高度估计研究提供了一个重要的补充选择。相比其他数据集,该数据集具有样本数量大、覆盖范围广、可获取性、建筑高度分布合理等特点,能够更好地满足深度学习网络的训练需求。在此基础上,本文采用相同的深度学习网络对BHDSI数据集及其他类似数据集进行了评估,并对比了多个网络使用BHDSI数据集时在建筑高度回归任务中的表现,深入分析了各网络的优劣。结果表明,与其他数据集相比,BHDSI数据集在建筑高度回归任务中的表现更加优异。进一步分析发现,使用BHDSI数据集时,建筑高度较低的区域其估计精度相对较高。此外,U-Net解码器用于建筑高度估计网络训练能够取得更高的精度。综上,BHDSI数据集为未来建筑高度估计领域的研究提供了重要的支持。展开更多
人工智能在信用风险评估中能有效识别风险并提升决策效率,然而,现有信用风险数据普遍存在类别不平衡问题,导致模型在预测时偏向多数类,影响评估的准确性和可靠性。针对数据不平衡问题,提出一种融合变分自编码器(VAE)和条件表格生成对抗...人工智能在信用风险评估中能有效识别风险并提升决策效率,然而,现有信用风险数据普遍存在类别不平衡问题,导致模型在预测时偏向多数类,影响评估的准确性和可靠性。针对数据不平衡问题,提出一种融合变分自编码器(VAE)和条件表格生成对抗网络(CTGAN)的混合生成模型(VCTGAN),用于合成高质量平衡数据集。通过VAE中的隐变量学习真实数据的关键特征和潜在分布,生成结构化隐变量作为原始CTGAN的输入;在数据生成器中引入自注意力机制用于更好地捕捉不平衡数据的突出特征;在判别器中加入对比损失模块来增强生成数据的类别间差异,达到提高生成数据质量的目的。通过在Taiwan Credit和Give Me Some Credit两个基准数据集上的系统实验验证,分别取得了89.91%和96.89%的最佳分类准确率,结果表明这种改进方法在处理信用数据不平衡方面明显优于传统方法。消融实验进一步验证了各组件对性能的贡献,证实了所提方法的合理性和有效性。它不仅生成高质量的平衡数据集,而且提高模型识别少数类别的能力,为解决金融领域的数据不平衡问题提供了新的技术方案。展开更多
Maize （Zea mays） is the most widely grown grain crop in the world, playing important roles in agriculture and industry. However, the functions of maize genes remain largely unknown. High-quality genome- wide transcr...Maize （Zea mays） is the most widely grown grain crop in the world, playing important roles in agriculture and industry. However, the functions of maize genes remain largely unknown. High-quality genome- wide transcriptome datasets provide important biological knowledge which has been widely and suc- cessfully used in plants not only by measuring gene expression levels but also by enabling co-expression analysis for predicting gene functions and modules related to agronomic traits. Recently, thousands of maize transcriptomic data are available across different inbred lines, development stages, tissues, and treatments, or even across different tissue sections and cell lines. Here, we integrated 701 transcriptomic and 108 epigenomic data and studied the different conditional networks with multi-dimensional omics levels. We constructed a searchable, integrative, one-stop online platform, the maize conditional co- expression network （MCENet） platform. MCENet provides 10 global/conditional co-expression net- works, 5 network accessional analysis toolkits （i.e., Network Search, Network Remodel, Module Finder, Network Comparison, and Dynamic Expression View） and multiple network functional support toolkits （e.g., motif and module enrichment analysis）. We hope that our database might help plant research communities to identify maize functional genes or modules that regulate important agronomic traits.展开更多
This paper tests various scenarios of feature selection and feature reduction, with the objective of building a real-time anomaly-based intrusion detection system. These scenarios are evaluated on the realistic Kyoto ...This paper tests various scenarios of feature selection and feature reduction, with the objective of building a real-time anomaly-based intrusion detection system. These scenarios are evaluated on the realistic Kyoto 2006+ dataset. The influence of reducing the number of features on the classification performance and the execution time is measured for each scenario. The so-called HVS feature selection technique detailed in this paper reveals many advantages in terms of consistency, classification performance and execution time.展开更多
Detecting the anomalous entity in real-time network traffic is a popular area of research in recent times.Very few researches have focused on creating malware that fools the intrusion detection system and this paper f...Detecting the anomalous entity in real-time network traffic is a popular area of research in recent times.Very few researches have focused on creating malware that fools the intrusion detection system and this paper focuses on this topic.We are using Deep Convolutional Generative Adversarial Networks(DCGAN)to trick the malware classifier to believe it is a normal entity.In this work,a new dataset is created to fool the Artificial Intelligence(AI)based malware detectors,and it consists of different types of attacks such as Denial of Service(DoS),scan 11,scan 44,botnet,spam,User Datagram Portal(UDP)scan,and ssh scan.The discriminator used in the DCGAN discriminates two different attack classes(anomaly and synthetic)and one normal class.The model collapse,instability,and vanishing gradient issues associated with the DCGAN are overcome using the proposed hybrid Aquila optimizer-based Mine blast harmony search algorithm(AO-MBHS).This algorithm helps the generator to create realistic malware samples to be undetected by the discriminator.The performance of the proposed methodology is evaluated using different performance metrics such as training time,detection rate,F-Score,loss function,Accuracy,False alarm rate,etc.The superiority of the hybrid AO-MBHS based DCGAN model is noticed when the detection rate is changed to 0 after the retraining method to make the defensive technique hard to be noticed by the malware detection system.The support vector machines(SVM)is used as the malicious traffic detection application and its True positive rate(TPR)goes from 80%to 0%after retraining the proposed model which shows the efficiency of the proposed model in hiding the samples.展开更多
Anomaly based approaches in network intrusion detection suffer from evaluation, comparison and deployment which originate from the scarcity of adequate publicly available network trace datasets. Also, publicly availab...Anomaly based approaches in network intrusion detection suffer from evaluation, comparison and deployment which originate from the scarcity of adequate publicly available network trace datasets. Also, publicly available datasets are either outdated or generated in a controlled environment. Due to the ubiquity of cloud computing environments in commercial and government internet services, there is a need to assess the impacts of network attacks in cloud data centers. To the best of our knowledge, there is no publicly available dataset which captures the normal and anomalous network traces in the interactions between cloud users and cloud data centers. In this paper, we present an experimental platform designed to represent a practical interaction between cloud users and cloud services and collect network traces resulting from this interaction to conduct anomaly detection. We use Amazon web services (AWS) platform for conducting our experiments.展开更多
This paper introduces a Convolutional Neural Network (CNN) model for Arabic Sign Language (AASL) recognition, using the AASL dataset. Recognizing the fundamental importance of communication for the hearing-impaired, e...This paper introduces a Convolutional Neural Network (CNN) model for Arabic Sign Language (AASL) recognition, using the AASL dataset. Recognizing the fundamental importance of communication for the hearing-impaired, especially within the Arabic-speaking deaf community, the study emphasizes the critical role of sign language recognition systems. The proposed methodology achieves outstanding accuracy, with the CNN model reaching 99.9% accuracy on the training set and a validation accuracy of 97.4%. This study not only establishes a high-accuracy AASL recognition model but also provides insights into effective dropout strategies. The achieved high accuracy rates position the proposed model as a significant advancement in the field, holding promise for improved communication accessibility for the Arabic-speaking deaf community.展开更多
文摘现有基于预训练语言模型(PLM)的作文自动评分(AES)方法偏向于直接使用从PLM提取的全局语义特征表示作文的质量,却忽略了作文质量与更细粒度特征关联关系的问题。聚焦于中文AES研究,从多种文本角度分析和评估作文质量,提出利用图神经网络(GNN)对作文的多尺度特征进行联合学习的中文AES方法。首先,利用GNN分别获取作文在句子级别和段落级别的篇章特征;然后,将这些篇章特征与作文的全局语义特征进行联合特征学习,实现对作文更精准的评分;最后,构建一个中文AES数据集,为中文AES研究提供数据基础。在所构建的数据集上的实验结果表明,所提方法在6个作文主题上的平均二次加权Kappa(QWK)系数相较于R2-BERT(Bidirectional Encoder Representations from Transformers model with Regression and Ranking)提升了1.1个百分点,验证了在AES任务中进行多尺度特征联合学习的有效性。同时,消融实验结果进一步表明了不同尺度的作文特征对评分效果的贡献。为了证明小模型在特定任务场景下的优越性,与当前流行的通用大语言模型GPT-3.5-turbo和DeepSeek-V3进行了对比。结果表明,使用所提方法的BERT(Bidirectional Encoder Representations from Transformers)模型在6个作文主题上的平均QWK比GPT-3.5-turbo和DeepSeek-V3分别高出了65.8和45.3个百分点,验证了大语言模型(LLMs)在面向领域的篇章级作文评分任务中,因缺乏大规模有监督微调数据而表现不佳的观点。
文摘利用光学和SAR遥感影像进行建筑高度估计对于理解城市形态和优化城市存量空间具有重要意义。然而,现有的数据集存在诸多局限:由于样本数量较少,难以满足基于深度学习的遥感信息提取需求,样本所覆盖的区域较为有限,无法提供足够的地理多样性和空间特征代表性,特别是针对中国区域的大规模建筑高度数据集尤为缺乏。此外,数据集的开源性不足,限制了其在更广泛的研究中的应用和验证。为解决这些问题,本文构建了一个面向深度学习的基于Sentinel影像的建筑物高度数据集BHDSI(Building Height Estimation Dataset Based on Sentinel Imagery),该数据集涵盖了中国62个城市的中心城区,共有5606个样本,覆盖了城市,农村等场景,是目前中国区域覆盖面积最大的建筑高度数据集。该数据集包含哨兵一号和哨兵二号的遥感影像以及建筑高度的真实值,样本大小是256×256,相比于64×64大小的数据集,为建筑高度估计研究提供了一个重要的补充选择。相比其他数据集,该数据集具有样本数量大、覆盖范围广、可获取性、建筑高度分布合理等特点,能够更好地满足深度学习网络的训练需求。在此基础上,本文采用相同的深度学习网络对BHDSI数据集及其他类似数据集进行了评估,并对比了多个网络使用BHDSI数据集时在建筑高度回归任务中的表现,深入分析了各网络的优劣。结果表明,与其他数据集相比,BHDSI数据集在建筑高度回归任务中的表现更加优异。进一步分析发现,使用BHDSI数据集时,建筑高度较低的区域其估计精度相对较高。此外,U-Net解码器用于建筑高度估计网络训练能够取得更高的精度。综上,BHDSI数据集为未来建筑高度估计领域的研究提供了重要的支持。
文摘人工智能在信用风险评估中能有效识别风险并提升决策效率,然而,现有信用风险数据普遍存在类别不平衡问题,导致模型在预测时偏向多数类,影响评估的准确性和可靠性。针对数据不平衡问题,提出一种融合变分自编码器(VAE)和条件表格生成对抗网络(CTGAN)的混合生成模型(VCTGAN),用于合成高质量平衡数据集。通过VAE中的隐变量学习真实数据的关键特征和潜在分布,生成结构化隐变量作为原始CTGAN的输入;在数据生成器中引入自注意力机制用于更好地捕捉不平衡数据的突出特征;在判别器中加入对比损失模块来增强生成数据的类别间差异,达到提高生成数据质量的目的。通过在Taiwan Credit和Give Me Some Credit两个基准数据集上的系统实验验证,分别取得了89.91%和96.89%的最佳分类准确率,结果表明这种改进方法在处理信用数据不平衡方面明显优于传统方法。消融实验进一步验证了各组件对性能的贡献,证实了所提方法的合理性和有效性。它不仅生成高质量的平衡数据集,而且提高模型识别少数类别的能力,为解决金融领域的数据不平衡问题提供了新的技术方案。
基金supported by the National Natural Science Foundation of China (Nos. 31771467, 31571360 and 31371291)
文摘Maize （Zea mays） is the most widely grown grain crop in the world, playing important roles in agriculture and industry. However, the functions of maize genes remain largely unknown. High-quality genome- wide transcriptome datasets provide important biological knowledge which has been widely and suc- cessfully used in plants not only by measuring gene expression levels but also by enabling co-expression analysis for predicting gene functions and modules related to agronomic traits. Recently, thousands of maize transcriptomic data are available across different inbred lines, development stages, tissues, and treatments, or even across different tissue sections and cell lines. Here, we integrated 701 transcriptomic and 108 epigenomic data and studied the different conditional networks with multi-dimensional omics levels. We constructed a searchable, integrative, one-stop online platform, the maize conditional co- expression network （MCENet） platform. MCENet provides 10 global/conditional co-expression net- works, 5 network accessional analysis toolkits （i.e., Network Search, Network Remodel, Module Finder, Network Comparison, and Dynamic Expression View） and multiple network functional support toolkits （e.g., motif and module enrichment analysis）. We hope that our database might help plant research communities to identify maize functional genes or modules that regulate important agronomic traits.
文摘This paper tests various scenarios of feature selection and feature reduction, with the objective of building a real-time anomaly-based intrusion detection system. These scenarios are evaluated on the realistic Kyoto 2006+ dataset. The influence of reducing the number of features on the classification performance and the execution time is measured for each scenario. The so-called HVS feature selection technique detailed in this paper reveals many advantages in terms of consistency, classification performance and execution time.
基金This project was funded by the Deanship of Scientific Research(DSR)at King Abdulaziz University,Jeddah,under Grant No.RG-91-611-42.
文摘Detecting the anomalous entity in real-time network traffic is a popular area of research in recent times.Very few researches have focused on creating malware that fools the intrusion detection system and this paper focuses on this topic.We are using Deep Convolutional Generative Adversarial Networks(DCGAN)to trick the malware classifier to believe it is a normal entity.In this work,a new dataset is created to fool the Artificial Intelligence(AI)based malware detectors,and it consists of different types of attacks such as Denial of Service(DoS),scan 11,scan 44,botnet,spam,User Datagram Portal(UDP)scan,and ssh scan.The discriminator used in the DCGAN discriminates two different attack classes(anomaly and synthetic)and one normal class.The model collapse,instability,and vanishing gradient issues associated with the DCGAN are overcome using the proposed hybrid Aquila optimizer-based Mine blast harmony search algorithm(AO-MBHS).This algorithm helps the generator to create realistic malware samples to be undetected by the discriminator.The performance of the proposed methodology is evaluated using different performance metrics such as training time,detection rate,F-Score,loss function,Accuracy,False alarm rate,etc.The superiority of the hybrid AO-MBHS based DCGAN model is noticed when the detection rate is changed to 0 after the retraining method to make the defensive technique hard to be noticed by the malware detection system.The support vector machines(SVM)is used as the malicious traffic detection application and its True positive rate(TPR)goes from 80%to 0%after retraining the proposed model which shows the efficiency of the proposed model in hiding the samples.
文摘Anomaly based approaches in network intrusion detection suffer from evaluation, comparison and deployment which originate from the scarcity of adequate publicly available network trace datasets. Also, publicly available datasets are either outdated or generated in a controlled environment. Due to the ubiquity of cloud computing environments in commercial and government internet services, there is a need to assess the impacts of network attacks in cloud data centers. To the best of our knowledge, there is no publicly available dataset which captures the normal and anomalous network traces in the interactions between cloud users and cloud data centers. In this paper, we present an experimental platform designed to represent a practical interaction between cloud users and cloud services and collect network traces resulting from this interaction to conduct anomaly detection. We use Amazon web services (AWS) platform for conducting our experiments.
文摘This paper introduces a Convolutional Neural Network (CNN) model for Arabic Sign Language (AASL) recognition, using the AASL dataset. Recognizing the fundamental importance of communication for the hearing-impaired, especially within the Arabic-speaking deaf community, the study emphasizes the critical role of sign language recognition systems. The proposed methodology achieves outstanding accuracy, with the CNN model reaching 99.9% accuracy on the training set and a validation accuracy of 97.4%. This study not only establishes a high-accuracy AASL recognition model but also provides insights into effective dropout strategies. The achieved high accuracy rates position the proposed model as a significant advancement in the field, holding promise for improved communication accessibility for the Arabic-speaking deaf community.
