Search

This paper addresses the pertinent object localization problem in deep convolutional neural networks by introducing a spatial fuzzy post-processing function which allows the smooth transition of object edges within individual pixel's neighborhood. We accomplish the task of semantic segmentation by first computing class weights as a means of avoiding class bias or imbalance training. Our proposed FuzzyNet runs a convolutional encoderdecoder network architecture with the following novel features: (i) It incorporates a new Global Context Spatial Module (GCSM) (ii) It exploits the atrous spatial pyramid structure for enriching the semantic encoding (iii) It incorporates the transfer of lower level features connected to higher levels with contextual spatial feature maps (iv) It effectively achieves an attention component with an extensive focus on objects of interest. Thus, the fusion of spatial fuzzy function enables normalization of intensity variation at different object boundaries, avoidance of poor localization and ultimately resulting in quality semantic segmentation. The evaluation of our proposed FuzzyNet model achieves improved performance with respect to the accuracy and object boundary refinement on the PASCAL VOC 2012 and CamVid benchmark datasets.

Skeleton-based action human recognition has drawn increasing attentions due to its properties of robustness and conciseness, while studies in recently years mostly have focused on extracting global motion features of skeleton but ignored the correlation among joints of local parts of skeleton. In this paper, we proposed a multi-stream network model based on local part joints motion features, our model focus on features extraction of local part joint motion and effect of fusion method on action recognition, utilizing LSTM and CNN structure a new network unit to grasp spatio-temporal information of joints in skeleton sequences. In order to explore distinctive motion modality of skeletal part, multi-stream mode is adopted and conducting effective recognition with weighted-score fusion. We evaluated our method on the NTU-RGB+D dataset, our result demonstrate a comparable performance of the proposed model in human action recognition.

Most existing image enhancement methods function like a black box, which cannot clearly reveal the procedure behind each image enhancement operation. To overcome this limitation, in this paper, we design a progressive image enhancement framework, which generates an expected "good" retouched image with a group of self-interpretable image filters under the guidance of an aesthetic assessment model. The introduced aesthetic network effectively alleviates the shortage of paired training samples by providing extra supervision, and eliminate the bias caused by human subjective preferences. The self-interpretable image filters designed in our image enhancement framework, make the overall image enhancing procedure easy-to-understand. Extensive experiments demonstrate the effectiveness of our proposed framework.

RBAC and ABAC are well-known access control models due to their least privileges and dynamic behavior respectively. They also have some drawbacks like RBAC is unable to provide dynamic behavior and flexibility as well as ABAC is unable to provide tight security and ease of management of permissions as the RBAC can do. In this paper, a hybrid access control model is proposed and developed that combines the strengths of both models. The proposed model implements the concept of roles between a user and the user's attributes as well as between the object and object attributes, in the ABAC system. The proposed model decreases the load of the administrator, provides least of privileges concept in ABAC due to the addition of roles. Authors also implemented the proposed model and discussed with respect to a case study.

We propose a concurrent Fully Convolutional Networks(CFCN) structure which contains three Fully Convolutional Networks(FCN). Gaussian filter, Mean filter and Median filter are chosen to pre-process the original multimodal MR images. Then, we fuse the results from three networks. Finally, a Fully Connected Conditional Random Field (Fully Connected CRF) is used to accomplish the post-processing, improving the model's ability of detecting minute structures. Our model was trained and evaluated on BRATS 2015 challenge dataset. The results show that our model provides promising segmentations.

Convolutional Neural Networks (CNNs) have been widely used and achieve amazing performance, typically at the cost of very expensive computation. Some methods accelerate the CNN training by distributed GPUs those deploying GPUs on multiple servers. Unfortunately, they need to transmit a large amount of data among servers, which leads to long data transmitting time and long GPU idle time. Towards this end, we propose a novel hybrid parallelism architecture named "Wheel" to accelerate the CNN training by reducing the transmitted data and fully using GPUs simultaneously. Specifically, Wheel first partitions the layers of a CNN into two kinds of modules: convolutional module and fully-connected module, and deploys them following the proposed hybrid parallelism. In this way, Wheel transmits only a few parameters of CNNs among different servers, and transmits most of the parameters within the same server. The time to transmit data is significantly reduced. Second, to fully run each GPU and reduce the idle time, Wheel devises an alternate strategy deploying multiple workers on each GPU. Once one worker is suspended for receiving data, another one in the same GPU starts to execute the computing task. The workers in each GPU run concurrently and repeatedly like Wheels. Experiments are conducted to show the outperformance of the proposed scheme over the state-of-the-art parallel approaches.

Relational inference is a crucial technique for knowledge base population. The central problem in the study of relational inference is to infer unknown relations between entities from the facts given in the knowledge bases. Two popular models have been put forth recently to solve this problem, which are the latent factor models and the random-walk models, respectively. However, each of them has their pros and cons, depending on their computational efficiency and inference accuracy. In this paper, we propose a hierarchical random-walk inference algorithm for relational learning in large scale graph-structured knowledge bases, which not only maintains the computational simplicity of the random-walk models, but also provides better inference accuracy than related works. The improvements come from two basic assumptions we proposed in this paper. Firstly, we assume that although a relation between two entities is syntactically directional, the information conveyed by this relation is equally shared between the connected entities, thus all of the relations are semantically bidirectional. Secondly, we assume that the topology structures of the relation-specific subgraphs in knowledge bases can be exploited to improve the performance of the random-walk based relational inference algorithms. The proposed algorithm and ideas are validated with numerical results on experimental data sampled from practical knowledge bases, and the results are compared to state-of-the-art approaches.

Fast and accurate identification of active recursive domain name servers (RDNS) is a fundamental step to evaluate security risk degrees of DNS systems. Much identification work have been proposed based on network traffic measurement technology. Even though identifying RDNS accurately, they waste huge network resources, and fail to obtain host activity and distinguish between direct and indirect RDNS. In this paper, we proposed an approach to identify direct and forward RDNS based on our three key insights on their request-response behaviors, and proposed an approach to identify indirect RDNS based on CNAME redirect behaviors. To work in high-speed backbone networks, we further proposed an online connectivity estimation algorithm to obtain estimated values used in our identification approaches. According to our experiments, we can identify RDNS with a high accuracy by selecting the reasonable thresholds. The accuracy of identifying direct and forward RDNS can reach 89%.The accuracy of identifying indirect RDNS can reach 90%.Moreover, our work is capable of real-time analyzing high speed backbone traffics.