Search

Dynamic routing networks, aimed at finding the best routing paths in the networks, have achieved significant improvements to neural networks in terms of accuracy and efficiency. In this paper, we see dynamic routing networks in a fresh light, formulating a routing method as a mapping from a sample space to a routing space. From the perspective of space mapping, prevalent methods of dynamic routing did not take into account how inference paths would be distributed in the routing space. Thus, we propose a novel method, termed CoDiNet, to model the relationship between a sample space and a routing space by regularizing the distribution of routing paths with the properties of consistency and diversity. Specifically, samples with similar semantics should be mapped into the same area in routing space, while those with dissimilar semantics should be mapped into different areas. Moreover, we design a customizable dynamic routing module, which can strike a balance between accuracy and efficiency. When deployed upon ResNet models, our method achieves higher performance and effectively reduces average computational cost on four widely used datasets.

RF power transfer is becoming a reliable solution to energy supplement of Internet of Things (IoT) in recent years, thanks to the emerging off-the-shelf wireless charging and sensing platforms. However, as a core component of IoT, sensor nodes mounted with these platforms can not work and harvest energy simultaneously, due to the low-manufacture-cost requirement. This leads to a new design challenge of optimally scheduling sensor nodes’ operation states: working or recharging, to achieve a desirable network utility. In our design, we first consider a single-hop special case of small-scale networks. We transform the operation state scheduling problem into a linear programming problem, and obtain an optimal analytical solution. Then a general case of large-scale multi-hop networks is investigated. The multi-hop operation state scheduling problem is proved to be NP-hard. We show that the spatiotemporal coupling caused by time-varying network topology makes the problem quite challenging. Based on Lyapunov optimization technique, we design a State Scheduling Algorithm (SSA) with a proved performance guarantee. Our algorithm decouples the primal problem by defining a dynamic energy threshold vector, which successfully schedules each sensor node to the desirable state according to its energy level. To verify our design, the SSA is implemented on a Powercast wireless charging and sensing testbed, achieving about 85 percent of the theoretical optimal with quite low time complexity. Furthermore, numerous simulation results demonstrate that the SSA outperforms the baseline algorithms and achieves good performance under different network settings.

With more users outsourcing their applications to the cloud, resource pricing becomes an important issue for IaaS cloud management. Jointly considering her own bidding budget and the price of cloud resources, each user is self-motivated to purchase cloud resources according to her resource demand which maximizes her own utility. Meanwhile, the cloud service provider (CSP) regulates the price of cloud resources with a certain profitability objective achieved. With an elaborate resource pricing strategy, the goals from users and the CSP are balanced and respectively satisfied to some extent. This article provides an insight into the market-oriented cloud pricing strategy. In specific, we propose an auction market in the IaaS cloud, where multiple users with heterogeneous bidding budgets and QoS requirements subscribe cloud resources according to their resource demands. The resource pricing and demand allocation scheme targeting revenue maximization also satisfies essential properties including budget feasibility, incentive compatibility and envy-freeness. To attack the NP-hardness and non-convexity of revenue maximization problem, we design a price-incentive resource auction mechanism namely RARM, which preserves an ( $1+\alpha $ ) approximation ratio on revenue maximization. Finally, we evaluate our RARM mechanism based on the real-world dataset to certify the efficacy of our proposed approach.

For a cloud service provider, it necessitates an emerging cloud ecosystem to consolidate the existing users and attract more potential users, further gaining its market share. Therefore, in this article, we design a price-incentive resource auction mechanism in cloud environment. In response to the cloud resource price, each user synthesizes her bidding budget and QoS requirement, and purchases cloud resources according to her resource demand in a strategic manner. The cloud service provider, meanwhile, can regulate the resource demands of users through conducting a market-based pricing strategy, against too low prices to cover the operational costs (i.e., energy costs) or too high prices resulting in user churn. In virtue of an elaborate market-based pricing strategy, the interests of users and the cloud service provider are balanced. Our price-incentive resource auction mechanism targets to stimulate maximum users willing to purchase resources and perform their applications at the cloud, on the premise of a minimum profit rate guaranteed for the cloud service provider. It is also able to provide budge balance and truthfulness guarantee, and satisfy the envy-freeness. In order to carry out the above objectives, we carefully design the user utility function reflecting the complicated user interest, and formulate our resource pricing and auction problem as a bin packing problem, which has non-polynomial computational complexity. Regarding the NP-hardness of optimization problem and the concavity of user utility, we present a computational-efficient ( $1+\epsilon $ )-approximate algorithm namely PIRA. Finally, we conduct simulations based on the real-world dataset to validate the effectiveness of our proposed approach.

High-capacity image steganography, aimed at concealing a secret image in a cover image, is a technique to preserve sensitive data, e.g., faces and fingerprints. Previous methods focus on the security during transmission and subsequently run a risk of privacy leakage after the restoration of secret images at the receiving end. To address this issue, we propose a framework, called Multitask Identity-Aware Image Steganography (MIAIS), to achieve direct recognition on container images without restoring secret images. The key issue of the direct recognition is to preserve identity information of secret images into container images and make container images look similar to cover images at the same time. Thus, we introduce a simple content loss to preserve the identity information, and design a minimax optimization to deal with the contradictory aspects. We demonstrate that the robustness results can be transferred across different cover datasets. In order to be flexible for the secret image restoration in some cases, we incorporate an optional restoration network into our method, providing a multitask framework. The experiments under the multitask scenario show the effectiveness of our framework compared with other visual information hiding methods and state-of-the-art high-capacity image steganography methods., Accepted to Transaction of Image Processing

Formulated as a conditional generation problem, face animation aims at synthesizing continuous face images from a single source image driven by a set of conditional face motion. Previous works mainly model the face motion as conditions with 1D or 2D representation ( e.g., action units, emotion codes, landmark ), which often leads to low-quality results in some complicated scenarios such as continuous generation and large-pose transformation. To tackle this problem, the conditions are supposed to meet two requirements, i.e., motion information preserving and geometric continuity. To this end, we propose a novel representation based on a 3D geometric flow, termed facial flow, to represent the natural motion of the human face at any pose. Compared with other previous conditions, the proposed facial flow well controls the continuous changes to the face. After that, in order to utilize the facial flow for face editing, we build a synthesis framework generating continuous images with conditional facial flows. To fully take advantage of the motion information of facial flows, a hierarchical conditional framework is designed to combine the extracted multi-scale appearance features from images and motion features from flows in a hierarchical manner. The framework then decodes multiple fused features back to images progressively. Experimental results demonstrate the effectiveness of our method compared to other state-of-the-art methods.

Customer churn as a common but extremely harmful phenomenon haunts service operators in telecommunications industry for a long time. Discriminating customers prone to churning in the early stage and taking precaution measures can mitigate customer churn effectively. Unlike previous works which mainly considered the inter-operator customer churn, we focus on a new problem of intra-operator customer churn, that is, customers abandon their fourth generation (4G) mobile communication services and switch to third generation/second generation (3G/2G). Due to quite different mechanisms and causes, previous studies cannot tackle this new problem well. In this paper, we propose a novel profit maximizing framework to solve this special customer churn problem. Specifically, we first build a profit maximizing classifier to predict whether a customer is prone to churn or not, in which profitable customers have more probability to be classified correctly. Then, we describe and quantify the switching behaviors of 4G customers by introducing a notion of switching score. Finally, we investigate relations between operators and customers, finding that some features of 4G service plans provided by operators indeed affect switching behaviors of 4G customers remarkably. This provides insight into reasonable design of 4G service plans for operators in the future. Our framework uncovers the root causes of intra-operator customer churn and solve the problem well. Experimental results based on real data demonstrate the effectiveness of our framework.

Wireless charging technologies have attracted much attention from both academia and industry due to their great potential of freeing devices from wires and batteries, becoming one of the key research areas in recent years. Many platforms have been designed based on different charging technologies for a wide range of scenarios, including structural monitoring, habitat monitoring, wearable devices, volcano monitoring, and so on. However, these existing platforms cannot be deployed quickly and flexibly in many harsh environments, resulting in high deployment cost. In this article, we propose Q-Charge, a wireless charging platform integrated with a quadcopter, which is able to provide desirable energy for large-scale sensing applications. Q-Charge is composed of three components: an energy transfer module, a programmable quadcopter unit, and a cloud server. A wireless charger is installed on a quadcopter to achieve remote energy delivery, and a sink node retrieves data packets from sensors and forwards them to the cloud via a service gateway. Our platform has been successfully implemented for building structure and farmland monitoring, which demonstrates Q-Charge to be effective and efficient.