Search

This work aims to address the image classification problem under open-set protocol: classes in test set do not appear in the training set. Intuitively, convolutional Neural Network (CNN) with softmax loss is a straight-forward solution. However, the unknown class (is not predefined in the training set) makes the boundaries of intra-class and inter-class more blurred, which brings more challenges for image classification. Although some softmax variants, such as center loss, CosFace loss etc., focus on learning discriminative features by minimizing the intra-class distance, they do not explicitly maximize inter-class distance, which is more important for open-set problem verified by our experiments. Besides, even though deep metric learning, such as with the contrastive loss and the triplet loss, can learn discriminative features of intra-class and inter-class, it needs a time-consuming image sampling process during training. In this paper, we propose a novel normalized maximal margin (NMM) loss for open-set image classification, which not only explicitly minimizes intra-class distance and maximizes inter-class distance, but also defines their margins. Specially, after analyzing the advantage of angular space that the softmax loss normalized by the feature and weights through geometric interpretation, we make NMM work in angular space. Then, the validity of NMM for discriminative features learning is demonstrated from the view of geometric interpretation as well. After that, we innovatively determine the upper bound of inter-class margin by theoretical analysis. Finally, extensive experiments are conducted on popular datasets: CIFAR-100 (object recognition), ImageNet (image classification), LFW (face recognition) and MSMT17 (person re-identification) to verify the effectiveness of NMM. The experimental results show that NMM achieves very competitive performance.

Periodically data plane update is a necessary task in Software Defined Networking(SDN). Most prior approaches update many flows without considering which flows should be updated. Hence, most of them get long update delay. However, due to the dramatic increase of data arriving in data centers in the coming 5G networks, flows to be updated should be carefully chosen. Otherwise, unsatisfactory update delay may occur, which will make the route after the reconfiguration invalid for most flows and even bring about a poor user experience. In this paper, we present a Timely Route Update Scheme(TRUS). Different from previously proposed update approaches, TRUS not only meets the real-time route update demand for data centers, but also guarantees bandwidth requirements of delay-sensitive flows. Further, it can avoid the congestion on “hot” links in the update process. Simulation results verify the effectiveness of our scheme and demonstrate it can largely cut down the route update latency while obtaining a good network performance when comparing with some former approaches.

Intrusion detection can identify unknown attacks from network traffics and has been an effective means of network security. Nowadays, existing methods for network anomaly detection are usually based on traditional machine learning models, such as KNN, SVM, etc. Although these methods can obtain some outstanding features, they get a relatively low accuracy and rely heavily on manual design of traffic features, which has been obsolete in the age of big data. To solve the problems of low accuracy and feature engineering in intrusion detection, a traffic anomaly detection model BAT is proposed. The BAT model combines BLSTM (Bidirectional Long Short-term memory) and attention mechanism. Attention mechanism is used to screen the network flow vector composed of packet vectors generated by the BLSTM model, which can obtain the key features for network traffic classification. In addition, we adopt multiple convolutional layers to capture the local features of traffic data. As multiple convolutional layers are used to process data samples, we refer BAT model as BAT-MC. The softmax classifier is used for network traffic classification. The proposed end-to-end model does not use any feature engineering skills and can automatically learn the key features of the hierarchy. It can well describe the network traffic behavior and improve the ability of anomaly detection effectively. We test our model on a public benchmark dataset, and the experimental results demonstrate our model has better performance than other comparison methods.

Natural disasters bring huge loss of life and property to human beings. Unmanned aerial vehicles (UAVs) own the advantages of high mobility, high flexibility, and rapid deployment, and are important equipment during post-disaster rescue. However, UAVs usually have restricted battery and computing power. They are not fit for performing compute-intensive tasks during rescue. Since there are widespread parking resources in a city, multiple parked vehicles working together to compute the applications from UAVs in a post-disaster rescue is investigated to ensure the quality of experience (QoE) of the UAVs. To execute uploaded task effectively, surviving parked vehicles within the monitoring range of an UAV are arranged into a cluster as much as possible. Then, the task execution cost is analyzed. Furthermore, a deep reinforcement learning (DRL)-based offloading policy is constructed, which interacts with the environment in an intelligent way to achieve optimization goals. The simulation experiments show that the proposed offloading scheme has a higher task completion rate and a lower task execution cost than other baselines schemes.

Actuators and sensors are two types of important nodes in wireless sensor and actuator networks (WSANs). Actuators consume a large amount of energy for driving themselves move, handling events, communicating with each other, and computation. Hence, the sensors and actuators all suffer from a serious energy-constrained problem in WSANs. Recently, many excellent energy recharging works are proposed for wireless sensor networks (WSNs). However, the wireless energy delivery issue for WSANs is rarely addressed. In this paper, by deeply analyzing the characteristics of WSANs, we discuss the wireless mobile recharging issue for WSANs and present a novel actuator failure avoidance online charging scheme (AFAC). To prevent as many sensors failure as possible while ensuring no actuator failure, AFAC always tries to select the node that not only makes no actuator failure but also brings about the least number of nonfunctional sensors as the next target. Furthermore, since charging the actuators takes a long time, maximum charging threshold is set for actuators to prevent sensor nodes failure in the network. The simulation results demonstrate that our scheme gets the lowest sensor failure ratio with low charging cost and charging delay in comparison with other wireless charging schemes for WSANs.

SnS, is a promising anode material for lithium ion batteries (LIBs) and sodium ion batteries (SIBs), however, undergoes poor cyclic lifespan due to its huge volume changes and bad electroconductivity. Here, a modified CVD method is used to directly grow graphene‐like carbon film on the surface of SnS nanosheet arrays which are supported by Co‐, N‐modified porous carbon fibers (CCF@SnS@G). In the strategy, the SnS nanosheet arrays confined into the integrated carbon matrix containing porous carbon fibers and graphene‐like carbon film, perform a greatly improved electrochemical performance. In situ TEM experiments reveal that the vertical graphene‐like carbon film can not only protect the SnS nanosheet from destruction well and enhance the conductivity, but also transforms SnS nanosheet into ultrafine nanoparticles to promote the electrochemical kinetics. Systematic electrochemical investigations exhibit that the CCF@SnS@G electrode delivers a stable reversible capacity of 529 mAh g−1 at a high current density of 5 A g−1 for LIBs and 541.4 mAh g−1 at 2 A g−1 for SIBs, suggesting its good potential for anode electrodes., SnS nanosheet arrays sandwiched into the integrated carbon matrix containing porous carbon fibers and graphene‐like carbon film are prepared and provide a great improvement in the electrochemical performance of lithium ion and sodium ion batteries.

In vehicular ad hoc networks (VANETs), message dissemination to one or more vehicles is very challenging, due to frequent network disconnections and uncertain locations of the destination vehicles. Deploying roadside units (RSUs) is a possible solution to overcome the challenges, but it often requires a large amount of investment. In this paper, we propose the idea of Parking Backbone, which does not need any RSUs but leverages a virtual overlay network formed by outside parked vehicles to track vehicles and to disseminate messages between moving vehicles. Our scheme consists of three parts. At first, to each road, parked vehicles both at roadside and off street are grouped into a cluster as far as possible. An urban overlay network is established based on this type of clusters for data transmission. Secondly, to a specific vehicle, a daily mobility model is established, to determine its location through a corresponding location prediction algorithm. Finally, a novel message delivery scheme is designed to efficiently transmit messages to destination vehicles through the proposed virtual overlay network. Thanks to the extensive and stable outside parking in cities, once grouped into the overlay structure, data transmission can be easily achieved over the Parking Backbone. Extensive simulation results prove that our scheme achieves high performance in message dissemination.

Information interaction is a crucial part of modern transportation activities. In this paper, we propose the idea of PASS: a parking-lot-assisted carpool method over vehicular ad hoc networks (VANETs). PASS aims at optimizing transport utilization by the carpooling among car drivers who cover a part of the same traveling route. With wireless device enabled in the vehicle, a user can easily get matched vehicles information and then express his travel demands via radio queries over VANETs to the corresponding driver. The driver can decide whether to provide carpooling services or not. We investigate the main challenges in PASS design, via the parking lot to collect vehicle trajectories, via accelerator sensor to sense vehicle's movement, establish a routing tree to delivery vehicle trajectory information to nearby parking lots, and design a suitable matching scheme to match the target vehicle in VANETs. Finally, simulation results prove that PASS is effective and efficient in carpooling among vehicle drivers.

Data dissemination is the fundamental operation in vehicular ad hoc networks (VANETs); for example, after an accident or congestion is detected by the corresponding sensors mounted on the vehicles, an alert message should be swiftly disseminated to the vehicles moving towards the affected areas. However, the unique characteristics of VANETs, such as high mobility of vehicle nodes, intermittent connectivity, and rapidly dynamic topology, make data dissemination over them extremely challenging. Motivated by the fact that there are large amounts of roadside parked vehicles in urban areas, this paper proposes a parking-based data dissemination scheme for VANETs. Data to be disseminated are buffered at the roadside parked vehicle, which continuously provides data dissemination services for the vehicles passing by. We analyze the challenging issues in achieving parking-based data dissemination and provide possible solution for each issue. Theoretical results illustrate the effectiveness of our approach, and simulation results based on a real city map and realistic traffic situations show that the proposed data dissemination paradigm achieves a higher delivery ratio with lower network load and reasonable delivery delay.