Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-11-29 Cooperative journals: 《计算机应用研究》
Abstract： In the background of information overload, this is challenging to mine and organize meaningful concepts and their semantic connections from a set of related documents under the same topic in open information extraction. Thus, this paper proposed a multi-document conceptual graph model based on open-domain information extraction. Firstly, documents were ranked according to the improved TF-IDF weight of extracted topic words under the predefined topics, then the model relayed on a serious of methods, including coreference resolution, weight computation, open-domain information extraction method to extract numerous representative subject-predicate-object triples from multiple documents. For filtering out the noise of open-domain information approach itself and improving the accuracy of information extraction, this paper presented a fact filtering algorithm to retain only the most salient, compatible facts as well as a form of multiple conceptual subgraphs. Finally, in combined with the equivalent concepts and relationships across different subgraphs to connect into a fully connected conceptual graph with expressive topic ability. Experiments on Signal Media dataset illustrated that the proposed model has the ability to discern and effectively group the key information corresponds to specific topics within and across documents, and formed conceptual graph outperforms state-of-the-art the algorithms in terms of the coverage rate of topic concepts as well as the compatible facts. Besides, this model also has the important significance for the automatic Abstract: on.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-06-19 Cooperative journals: 《计算机应用研究》
Abstract： Considering the security risk of data stored in cloud and the deficiency of existing data integrity checking schemes, this paper proposed a multiple-replica remote data integrity checking protocol with efficient data dynamic update in cloud storage. The scheme was suitable for multiple replica scenario, and achieved multiple-replica provable data possession at a small cost on the basis of existing cloud data integrity checking schemes. By introducing authenticated data structure called rank-based MHT, the protocol supported full data dynamic operations. Through replica correlation, multiple replicas could be updated synchronously. Security analysis and experiment results show the security and effectiveness of the proposed protocol, and can guarantee the privacy of multiple replicas at the same time.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-05-24 Cooperative journals: 《计算机应用研究》
Abstract： A dominating set D⊆V of a graph G=(V,E) is a subset of vertices, such that every vertex of G is either in D, or adjacent to at least one vertex in D. The connected dominating set problem asks to find a dominating set S with minimum number of vertices and the induced subgraph G[S] of S is connected. The connected dominating set problem is a classical NP-hard problem, which could be applied to connected facility location, ad-hoc networks and many other areas. For the connected dominating set problem in undirected graphs, this paper carefully analyzes the structural properties, explores a number of effective reduction rules as well as branching rules and provides a branch-and-search algorithm. A measure-and-conquer method is also introduced to analyze the running time bound. Finally, an exact algorithm with a running time complexity of O*(1.93^n) is obtained.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-05-20 Cooperative journals: 《计算机应用研究》
Abstract： Voiceprint recognition was one of the most popular biometric identification technologies, which could identify a speaker based on his voice. This paper proposed CDRNN, a voiceprint recognition scheme. CDRNN combined CNN and Deep RNN into a unified model and took advantages of both of them. For CNN was good at extracting characteristics from images, it could generate several spectrograms based on the original voice signal at first. And then, CNN would extract unique features from these spectrograms. . Finally, Deep RNN would output the speaker's identification based on these unique features. Simulation results show that CDRNN performs better than GMM-UBM and DNN-based approach.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-05-18 Cooperative journals: 《计算机应用研究》
Abstract： Cloud storage system often adopts replica technique to guarantee availability and reliability, and replica placement is a key issue. In order to solve the problem of high access overhead to replicas, this paper proposed a replica placement algorithm based on discrete glowworm swarm optimization. Through mathematical mode establishment for user access overhead, this algorithm computed the fitness function of glowworm position, updated individual position and then obtained appropriate nodes for replica placement. It conducted several simulations and compared with the replica placement strategy based on ant algorithm. The results show that the proposed algorithm can select appropriate nodes for replica placement, have a better convergence and reduce the replica access overhead.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-04-24 Cooperative journals: 《计算机应用研究》
Abstract： Current salient object detection algorithms based on deep neural network (DNN) are usually not able to be aware of the structure of instance, making the generated saliency maps fail to cover the entire salient object region, and thus drag down the accuracy. To solve this problem, we introduced a novel multi-stream deep neural network, in which four components were integrated in a single framework: feature extractor, object skeleton sub-network, salient object sub-network and cross-domain connections. Firstly, during the learning and testing process, the salient object detection sub-network encoded the object structure which was extracted by using object skeleton detection sub-network through the cross-domain connections, so as to make the deep model be aware of the information of object structure and overcome the problem of incomplete detection of the target area. Then, to further improve the accuracy, we proposed to use a dense conditional random field based algorithm as the refinement post-process, so as to generate a more accurate saliency map as the final results. Experimental evaluations were conducted on three widely-used benchmarks and the results show that the proposed algorithm outperforms all existing DNN-based detection algorithms in accuracy and efficiency. This also indicates that integrating object structure information into deep neural network model is meaningful, which can help to improve the overall accuracy.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-04-17 Cooperative journals: 《计算机应用研究》
Abstract： Ambient occlusion is widely used in real time rendering for computing visibility of ambient light, especially in computer game and visualization domains. Although existing screen-space solutions can provide real time performance, they cannot generate accurate results and preserve details. This paper proposed a new solution based on Monte-Carlo ray tracing. To achieve real time performance, this method sampled the visibility of each pixel with sparse rays and then denoised the ray tracing results with a convolutional neural network (CNN) . To meet the real-time requirement of ambient occlusion computing, the algorithm further improved and optimized the network structure. Experimental results show that proposed method not only achieves the real time performance but also produces more accurate and detailed results. 牋
Hits
Hits
Downloads
Comment