Search

Radio Frequency IDentification (RFID) technology is widely used in the inventory management of goods. Fast searching a subset consisting of some particular target tags is of practical importance for a variety of applications. Existing protocols are inefficient due to the low utilization of frame. In this paper, in order to improve the utilization of slot frame, we first generalize Cuckoo hash to extended Cuckoo hash which can use more than two hash functions. Then, we propose an extended CUckoo hash-based missing tag Detection (CUD) protocol, which can efficiently identify all missing tags within a predefined specific set. In CUD, the reader first uses the extended cuckoo hash to generate a fingerprint vector of the target tags. By using this vector, almost all target tags can be assigned a single slot and the majority of non-target tags can be filtered out. Furthermore, we optimize the parameters through theoretical analysis to minimize the execution time of the protocol. Extensive simulation results demonstrate that CUD is superior to the state-of-the-art protocols.

When people search for flight ticket online, they interact with a web site by selecting the ticket information from a set of drop-down menus. This paper proposes a scheme that is more natural for people to interact with the flight ticket server. The scheme integrates a natural language interface with a flight ticket searching system. It allows the users to interact with the flight searching system using natural language. The system is based on the BERT model. Compared with the existing scheme, the proposed scheme achieves good accuracy.

Discovering Frequent itemsets over uncertain databases has become a hot topic in data mining community due to the wide application of uncertain data. Several algorithms are proposed to mine frequent itemsets over uncertain data. However, the existing mining algorithms generally suffer from extensive computation and prohibitively running time when the uncertain data is large or the frequent threshold is low. To address this issue, we propose a novel expected support-based frequent itemsets (E-FIs) mining method which discovers approximate E-FIs from a random sample of the original uncertain dataset. To obtain high quality approximation, we employ the solid theory of the Vapnik-Chervonenkis dimension. The mining result is thus a superset of the E-FIs with high probability guarantee. Finally, we conduct experiments on two public databases and evaluate our method in terms of accuracy, running time and scalability.

With the development of Internet of things, more and more personal information needs to be collected in the process of applying Internet of things technology, which brings about the problem of data privacy protection. The algorithm proposed in this paper uses the differential privacy model to process the information network collected by the wireless sensor network terminal in the perception layer of the Internet of things and transmits it to the network layer to achieve the purpose of information protection. In order to reduce the added differential privacy noise, this paper firstly makes dynamic community discovery of the network graph, and adds more noise to the edge of nodes within the community. Then the hierarchical random graph model is used to represent the network graph, and the noise is added to the connection probability between the nodes in the graph, instead of adding noise to the edge directly. Considering the dynamic change of the information collected by the wireless sensor network, this paper proposes a time window partition and dynamic network community discovery algorithm, and publishes a purification map for each time window. Through hierarchical sampling, the time cost is reduced and the cumulative error is reduced, while the network structure characteristics of the time window are maintained as far as possible. Theoretical analysis and experiments show that the algorithm proposed in this paper can preserve the important network structure characteristics of the original graph under the premise of satisfying the differential privacy protection model.

Differential privacy is a security guarantee model which widely used in privacy preserving data publishing, but the query result can’t be used in data research directly, especially in high-dimensional datasets. To address this problem, we propose a dimensionality reduction method. The core idea of this method is using a series of low-dimensional datasets to reconstruct a high-dimensional dataset, it improves data availability eventually. The main issue of this method is the reconstruction integrity, so a special sampling via set cover model is proposed in this article, which builds a multidimensional composite marginal tables set as a new middleware in differential privacy model. As a result, any form of disjunctive queries can be answered, and the accuracy of data query is improved. The experiment results also show the effectiveness of our method in practice.

We present a novel 2D shape decomposition algorithm via a recursive partitioning process. Starting with the contour points of a shape, we repeatedly separate the points into two parts by spectral clustering, until the stopping condition is met. Motivated by the fact that the points in a convex part are mutually visible, we regard the visibility matrix of points as the affinity matrix of spectral clustering to obtain a near-convex decomposition. Additionally, we present an efficient stopping rule to avoid over-segmentation on the shape branches. The stopping criterion is based on a novel shape signature called visible protrusion strength which can be used to measure the segmentability of a sub-shape. Finally, we demonstrate the efficiency of our algorithm on a variety of publicly available shapes, and provide qualitative and quantitative comparisons with state-of-art approaches.

Salient object detection aims to identify the most visually distinctive objects or regions in an image. Bottom-up methods always attract widespread attention in salient object detection. However, some bottom-up methods which based on background prior tend to outstand the salient object's edges rather than uniformly propagating the saliency value to the interior. In this paper, we present a Robust Background Detection Optimization (RBDO), which effectively improves an existing robust background detection method. This new method aims at the interior of salient objects which have uniform saliency values. First of all, a foreground probability map is constructed based on the background detection with intuitive geometrical interpretation. According to our proposed strategy, we get this point of the foreground probability map as a center point. Secondly, we combine center prior based on global contrast with the foreground probability, by which the accuracy of the foreground probability can be improved. New foreground probability is converted to background probability, and a weak saliency map can be obtained by background weighted contrast. Finally, we combine saliency optimization method with graph cut method to optimize saliency map together. Extensive experiments demonstrate that our algorithm performs favorably against state-of-the-art saliency detection methods on the four benchmark data sets.

Traffic behavior is closely related to the scene structure. Insufficient training samples will result in the problem of inappropriate scene modeling. An attribute-based transfer learning method for behavior recognition is put forward, which contains two main components: S-ATM Attribute Topic Model and S-CTM classifier model. S-ATM Attribute Topic Model is obtained by combined with the sources and sinks on the basis of the ATM Model, which realize modeling between behavioral properties and the underlying trajectory characteristics to learn the model parameters. Then the parameters of S-ATM Attribute Topic Model transfer to the S-CTM classifier model as attribute prior knowledge to improve the performance of target behavior classifier model. Experiments show that the proposed approach can transfer attribute priors from source categories to target categories effectively and significantly improve the recognition accuracy rate of the target behavior classifier model in zero-shot learning task.

Recently, saliency detection has attracted a great deal of research interests in the field of computer vision. Image contrast based on low-level feature is an efficient method for extracting salient regions. There are two classical algorithms based on the contrast method, which are the center prior and the background prior. However, both of center prior and background prior will lead to inaccurate extraction respectively when significant object is not in the center of picture or touches picture boundary. In order to extract the salient area accurately, a different method is proposed. Firstly, we divide original image into superpixels by Simple Linear Iterative Clustering (SLIC) algorithm, which maximize the salient areas maintain original shape and distinguish foreground and background roughly. Secondly, both background prior and center prior are improved respectively, which can solve significant object migration problem. Thirdly, we optimize the saliency map by the relationship between each superpixel and other neighbour superpixels, and then make a further distinction between background and foreground. So as to verify the efficiency of the proposed algorithm, we use four datasets, namely, MSRA-1000, ECSSD, SED, and SOD datasets in our experiments. Both qualitative and quantitative evaluations on four publicly available datasets demonstrate the robustness and efficiency of our proposed method against seven state-of-the-art methods.

Nowadays, hashing methods are widely used in large-scale approximate nearest neighbor search due to its efficient storage and fast retrieval speed. By these methods, the original data is usually hashed into binary codes which enables to measure the similarity by Hamming distance. When dealing with large-scale data, their binary codes can be used as direct indices in a hash table. However, codes longer than 32 bits are obviously not efficient. For the given binary codes, this paper proposes a multi-index hashing structure based on binary code substrings partitioning. Since substrings partitioning is essentially a combinatorial optimization problem, we propose a hierarchical and recursive partitioning approach to obtain an approximate solution to it. Furthermore, we adopt a query-adaptive fine-grained ranking approach in the neighbor search stage to alleviate the imbalance between multi-index tables. Finally, extensive experiments are conducted on two datasets MNIST and CIFAR-10, demonstrating that our method achieves state-of-the-art performance in terms of efficiency, precision and recall rate.

Recently a variety of approaches have been applied in protecting information privacy in Internet of Things systems such as intelligent security protection systems, smart home systems, or digital access control systems. In this paper, we study a fatal problem: how to remain secure in common sensing layer of cloud based IOT systems. For example, how to hide biological information such as fingerprint, facial expression, and body gestures, which are reversible to keep users' privacy still remains challengeable. We propose CLOAK which is a novel light-weight approach to ensure the privacy security in cloud based IOT systems. CLOAK takes Least Significant Bits (LSB) steganography to hide the vital information in fingerprint image, and it takes chaotic sequence encryption to resolve the information. CLOAK can resists static attack and protects the real users' privacy in terms of information security and integrity. Experimental and theoretical results demonstrate the effectiveness and security of the CLOAK.

The rapid development of Internet has resulted in massive information overloading recently. These information is usually represented by high-dimensional feature vectors in many related applications such as recognition, classification and retrieval. These applications usually need efficient indexing and search methods for such large-scale and high-dimensional database, which typically is a challenging task. Some efforts have been made and solved this problem to some extent. However, most of them are implemented in a single machine, which is not suitable to handle large-scale database.In this paper, we present a novel data index structure and nearest neighbor search algorithm implemented on Apache Spark. We impose a grid on the database and index data by non-empty grid cells. This grid-based index structure is simple and easy to be implemented in parallel. Moreover, we propose to build a scalable KNN graph on the grids, which increase the efficiency of this index structure by a low cost in parallel implementation. Finally, experiments are conducted in both public databases and synthetic databases, showing that the proposed methods achieve overall high performance in both efficiency and accuracy.

Nowadays many service providers accumulate large amounts of data. Sharing data can maximize its value, but the privacy may be leaked. Thus, the issue of privacy protection in data release has attracted the attention of academia and industry. Differential privacy has been applied to varied fields due to its excellent performance. The goal of this paper is to improve the accuracy of the published data while satisfying differential privacy. We propose an algorithm based on the hierarchical random graph (HRG) model. Considering that the users are usually interested in the data within certain communities, we apply differential privacy to the edges of a graph to various extents with the aid of community detection. Based on divide and conquer, our algorithm constructs an HRG for each community and merges sub-level HRG's into a complete one, which eliminates the inefficiency of the original algorithm when processing large-scale graphs. We treat different parts of a graph in different ways, keeping a balance between privacy and usability. According to several criteria in the experiments, our algorithm keeps the properties of the original graphs better.

As large-scale social networks flourish, influence maximization problem has become a hot topic in recent years. Existing studies consider only positive relationships in social networks. However, it is completely ignored that both positive and negative relationships may exist at the same time in many Online Social Networks (OSNs), e.g. Epinions and Slashdot. In this work, we propose an efficient influence maximization algorithm, namely Hybrid Potential-influence Greedy algorithm with Negative opinions (HPG-N) that uses the advantages of cumulative feature for influence maximization in both Linear Threshold model (LT) and its extension LT-N that incorporates negative opinion diffusion. We aim at solving the problem of influence maximization in social networks considering both positive and negative relationships with greater diffusion scope and more effectiveness than other algorithms. The algorithm we proposed can solve the problem of maximizing the positive influence spread effectively in real-world networks with both positive and negative relationships. In the heuristic stage, to seek seeds more quickly, we seek for those nodes, which can provide the greatest potential influence as seeds. In the greedy stage, we improve the traditional greedy algorithm to expand the scope of influence as much as possible. Experimental results show that our algorithm significantly performs better than baseline algorithms in the aspect of running time.

Recently, privacy preservation in publishing social network has become one of the most notable challenges. Researchers have developed a variety of methods to protect individual's privacy while maintaining utility at the same time. There exists such a situation that an adversary may identify the privacy of a victim with the background knowledge of public neighborhoods. Unfortunately, most of previous researches only focus on unclassified neighborhood attacks and ignore that public neighborhoods whose information is completely open will bring greater risk than private neighborhoods. In view of the availability of users' information, we adopt k-anonymity which belongs to graph modification methods to protect private users from public neighborhood attacks. Although k-anonymity could defend users from re-identification, a group of nodes may share the same sensitive labels which may be exploited by attackers to speculate private information. So we adopt l-diversity to guard the sensitive labels. We conduct our experiments in some social networks, and the results show that our method is effective.

Efficient scheduling of computing resources is a fundamental issue that the cloud computing needs to solve, which involves highly challenging load-balance of multiple computing resources. In order to achieve optimal balance between the executing speed, the average response time and the overall system utilisation during cloud computing resource allocation, a cloud computing resources scheduling optimisation algorithm is proposed based on wavelet perturbation-based bat algorithm (WPBA). The algorithm first employs wavelet perturbation to enhance Bat algorithm's performance followed by population-entropy-guided substitution to control diversity and improve the converging speed and accuracy. Then it adopts WPBA to achieve resources scheduling optimisation of the cloud computing. The experiment shows that using WPBA has significantly improved the overall performance of the algorithm and has also remarkably optimised the resource scheduling capability of cloud computing and heightened the overall resource utilisation.