Based on the framework of support vector machines （SVM） using one-against-one （OAO） strategy, a new multi-class kernel method based on directed aeyclie graph （DAG） and probabilistic distance is proposed to raise...Based on the framework of support vector machines （SVM） using one-against-one （OAO） strategy, a new multi-class kernel method based on directed aeyclie graph （DAG） and probabilistic distance is proposed to raise the multi-class classification accuracies. The topology structure of DAG is constructed by rearranging the nodes＇ sequence in the graph. DAG is equivalent to guided operating SVM on a list, and the classification performance depends on the nodes＇ sequence in the graph. Jeffries-Matusita distance （JMD） is introduced to estimate the separability of each class, and the implementation list is initialized with all classes organized according to certain sequence in the list. To testify the effectiveness of the proposed method, numerical analysis is conducted on UCI data and hyperspectral data. Meanwhile, comparative studies using standard OAO and DAG classification methods are also conducted and the results illustrate better performance and higher accuracy of the orooosed JMD-DAG method.展开更多
To enhance network security,this study employs a deep graph matching model for vulnerability similarity detection.The model utilizes a Word Embedding layer to vectorize data words,an Image Embedding layer to vectorize...To enhance network security,this study employs a deep graph matching model for vulnerability similarity detection.The model utilizes a Word Embedding layer to vectorize data words,an Image Embedding layer to vectorize data graphs,and an LSTM layer to extract the associations between word and graph vectors.A Dropout layer is applied to randomly deactivate neurons in the LSTM layer,while a Softmax layer maps the LSTM analysis results.Finally,a fully connected layer outputs the detection results with a dimension of 1.Experimental results demonstrate that the AUC of the deep graph matching vulnerability similarity detection model is 0.9721,indicating good stability.The similarity scores for vulnerabilities such as memory leaks,buffer overflows,and targeted attacks are close to 1,showing significant similarity.In contrast,the similarity scores for vulnerabilities like out-of-bounds memory access and logical design flaws are less than 0.4,indicating good similarity detection performance.The model’s evaluation metrics are all above 97%,with high detection accuracy,which is beneficial for improving network security.展开更多
聚类是大规模高维向量数据分析的关键技术之一.近年来,基于密度的聚类算法DBSCAN(density-based spatial clustering of applications with noise)因其无须预先指定聚类数量、能够发现复杂聚类结构并有效识别噪声点的特性,在数据分析领...聚类是大规模高维向量数据分析的关键技术之一.近年来,基于密度的聚类算法DBSCAN(density-based spatial clustering of applications with noise)因其无须预先指定聚类数量、能够发现复杂聚类结构并有效识别噪声点的特性,在数据分析领域得到了广泛应用.然而,现有的基于密度的聚类算法在处理高维向量数据时将产生极高的时间代价且面临维度灾难等问题,难以在实际场景中部署应用.此外,随着信息技术的发展,高维向量数据规模急剧增加,使用CPU进行高维向量聚类在时间代价和可扩展性等方面将面临更大的挑战.为此,提出一种GPU加速的高维向量聚类算法,通过引入K近邻(K-nearest neighbor,KNN)图索引加速DBSCAN的计算.首先,设计了GPU加速的并行K近邻图构建算法,显著降低了K近邻图索引的构建开销.其次,提出了基于层间并行的K-means树分区算法及基于广度优先搜索和核心近邻图的并行聚类算法,改进了DBSCAN算法的计算流程,实现了高并发向量聚类.最后,在真实向量数据集上进行了大量实验,并将所提出的方法与现有方法进行了性能对比.实验结果表明,所提方法在保证聚类精度的前提下,将大规模向量聚类的效率提高了5.7–2822.5倍.展开更多
Support Vector Clustering (SVC) is a kernel-based unsupervised learning clustering method. The main drawback of SVC is its high computational complexity in getting the adjacency matrix describing the connectivity for ...Support Vector Clustering (SVC) is a kernel-based unsupervised learning clustering method. The main drawback of SVC is its high computational complexity in getting the adjacency matrix describing the connectivity for each pairs of points. Based on the proximity graph model [3], the Euclidean distance in Hilbert space is calculated using a Gaussian kernel, which is the right criterion to generate a minimum spanning tree using Kruskal's algorithm. Then the connectivity estimation is lowered by only checking the linkages between the edges that construct the main stem of the MST (Minimum Spanning Tree), in which the non-compatibility degree is originally defined to support the edge selection during linkage estimations. This new approach is experimentally analyzed. The results show that the revised algorithm has a better performance than the proximity graph model with faster speed, optimized clustering quality and strong ability to noise suppression, which makes SVC scalable to large data sets.展开更多
【目的/意义】利用图书文本内容实现相似图书推荐,海量图书数据环境下提高图书相似度计算效率。【方法/过程】构建了一种基于图结构的相似图书内容推荐方法,在图书的文本内容进行短语抽取后计算短语网络中的TextRank值获得图书关键词,...【目的/意义】利用图书文本内容实现相似图书推荐,海量图书数据环境下提高图书相似度计算效率。【方法/过程】构建了一种基于图结构的相似图书内容推荐方法,在图书的文本内容进行短语抽取后计算短语网络中的TextRank值获得图书关键词,进而建立图书向量并结合层次可导航小世界算法(Hierarchcal Navigable Small World,HNSW)得到目标图书和推荐图书之间的相似度。【结果/结论】利用基于内容的相似图书推荐方法得到的用户评价平均准确率达到0.807,客观平均准确率显著高于TF-IDF和TextRank的文本表示方法,可以实现较好的图书推荐效果,HNSW算法将计算效率缩小到对数级别,对大数据环境下的相似图书计算效率起到一定的优化作用。【创新/局限】本研究创新性地结合图结构和HNSW算法提高了图书推荐的准确性和计算效率,但受限于对腾讯词典的依赖,影响了向量表达的普适性和跨语言适应性。展开更多
基金Sponsored by the National Natural Science Foundation of China(Grant No.61201310)the Fundamental Research Funds for the Central Universities(Grant No.HIT.NSRIF.201160)the China Postdoctoral Science Foundation(Grant No.20110491067)
文摘Based on the framework of support vector machines （SVM） using one-against-one （OAO） strategy, a new multi-class kernel method based on directed aeyclie graph （DAG） and probabilistic distance is proposed to raise the multi-class classification accuracies. The topology structure of DAG is constructed by rearranging the nodes＇ sequence in the graph. DAG is equivalent to guided operating SVM on a list, and the classification performance depends on the nodes＇ sequence in the graph. Jeffries-Matusita distance （JMD） is introduced to estimate the separability of each class, and the implementation list is initialized with all classes organized according to certain sequence in the list. To testify the effectiveness of the proposed method, numerical analysis is conducted on UCI data and hyperspectral data. Meanwhile, comparative studies using standard OAO and DAG classification methods are also conducted and the results illustrate better performance and higher accuracy of the orooosed JMD-DAG method.
基金Special Project Funded by Tsinghua University Press:“Engineering Drawing and CAD”Course Construction and Textbook Development。
文摘To enhance network security,this study employs a deep graph matching model for vulnerability similarity detection.The model utilizes a Word Embedding layer to vectorize data words,an Image Embedding layer to vectorize data graphs,and an LSTM layer to extract the associations between word and graph vectors.A Dropout layer is applied to randomly deactivate neurons in the LSTM layer,while a Softmax layer maps the LSTM analysis results.Finally,a fully connected layer outputs the detection results with a dimension of 1.Experimental results demonstrate that the AUC of the deep graph matching vulnerability similarity detection model is 0.9721,indicating good stability.The similarity scores for vulnerabilities such as memory leaks,buffer overflows,and targeted attacks are close to 1,showing significant similarity.In contrast,the similarity scores for vulnerabilities like out-of-bounds memory access and logical design flaws are less than 0.4,indicating good similarity detection performance.The model’s evaluation metrics are all above 97%,with high detection accuracy,which is beneficial for improving network security.
文摘聚类是大规模高维向量数据分析的关键技术之一.近年来,基于密度的聚类算法DBSCAN(density-based spatial clustering of applications with noise)因其无须预先指定聚类数量、能够发现复杂聚类结构并有效识别噪声点的特性,在数据分析领域得到了广泛应用.然而,现有的基于密度的聚类算法在处理高维向量数据时将产生极高的时间代价且面临维度灾难等问题,难以在实际场景中部署应用.此外,随着信息技术的发展,高维向量数据规模急剧增加,使用CPU进行高维向量聚类在时间代价和可扩展性等方面将面临更大的挑战.为此,提出一种GPU加速的高维向量聚类算法,通过引入K近邻(K-nearest neighbor,KNN)图索引加速DBSCAN的计算.首先,设计了GPU加速的并行K近邻图构建算法,显著降低了K近邻图索引的构建开销.其次,提出了基于层间并行的K-means树分区算法及基于广度优先搜索和核心近邻图的并行聚类算法,改进了DBSCAN算法的计算流程,实现了高并发向量聚类.最后,在真实向量数据集上进行了大量实验,并将所提出的方法与现有方法进行了性能对比.实验结果表明,所提方法在保证聚类精度的前提下,将大规模向量聚类的效率提高了5.7–2822.5倍.
基金TheNationalHighTechnologyResearchandDevelopmentProgramofChina (No .86 3 5 11 930 0 0 9)
文摘Support Vector Clustering (SVC) is a kernel-based unsupervised learning clustering method. The main drawback of SVC is its high computational complexity in getting the adjacency matrix describing the connectivity for each pairs of points. Based on the proximity graph model [3], the Euclidean distance in Hilbert space is calculated using a Gaussian kernel, which is the right criterion to generate a minimum spanning tree using Kruskal's algorithm. Then the connectivity estimation is lowered by only checking the linkages between the edges that construct the main stem of the MST (Minimum Spanning Tree), in which the non-compatibility degree is originally defined to support the edge selection during linkage estimations. This new approach is experimentally analyzed. The results show that the revised algorithm has a better performance than the proximity graph model with faster speed, optimized clustering quality and strong ability to noise suppression, which makes SVC scalable to large data sets.
文摘【目的/意义】利用图书文本内容实现相似图书推荐,海量图书数据环境下提高图书相似度计算效率。【方法/过程】构建了一种基于图结构的相似图书内容推荐方法,在图书的文本内容进行短语抽取后计算短语网络中的TextRank值获得图书关键词,进而建立图书向量并结合层次可导航小世界算法(Hierarchcal Navigable Small World,HNSW)得到目标图书和推荐图书之间的相似度。【结果/结论】利用基于内容的相似图书推荐方法得到的用户评价平均准确率达到0.807,客观平均准确率显著高于TF-IDF和TextRank的文本表示方法,可以实现较好的图书推荐效果,HNSW算法将计算效率缩小到对数级别,对大数据环境下的相似图书计算效率起到一定的优化作用。【创新/局限】本研究创新性地结合图结构和HNSW算法提高了图书推荐的准确性和计算效率,但受限于对腾讯词典的依赖,影响了向量表达的普适性和跨语言适应性。
