The goal of this paper is to manipulate image attributes using text description. Although many methods can synthesize images with new properties from text, they cannot fully preserve the text-independent content of the original image. There are two major limitations: (1) Some important details in image sub-region will be lost in the process of image modification; (2) Compared with the original image, the shape and edge of the object in the modified image will be more blurred. Therefore, we propose a novel framework edge aware generative adversarial network (EA-GAN) that uses edge information to guide image modification, which ensures the network’s ability to identify local regions and realizes accurate modification of image sub-regions. At the same time, an edge reconstruction loss (ERLoss) is added to the generator to constrain the generation of edges, generate sharper edges, and improve the clarity of the image. The experimental data on the CUB and Oxford-102 datasets show that the algorithm used in this paper can well distinguish the corresponding image features in the conditional text, and modify the image attribute of specific regions in the image.
Network Function Virtualization (NFV) has becoming an emerging technology for ensuring the reliability and security of data flows. The virtual network function (VNF) embedding problem, which tries to minimize the embedding cost has attracted extensive interests recently. However, the existing works always assume the fixed execution order of VNFs, which limits their application. Thus, we investigate the VNF embedding problem without such limitations in this paper. Firstly, a general transformation framework is proposed for the NFV-enabled unicast routing with adjustable order. Then, an optimal algorithm is proposed for the unicast VNF embedding without delay constraints. Additionally, an efficient algorithm is also proposed for such problem with delay constraints. Finally, we evaluate the proposed algorithms via numerical results, which show that our algorithms significantly outperform the existing benchmarks.
The lower limb exoskeleton robot system is one of the significant tools for the rehabilitation of patients with knee arthritis, which helps to enhance the health of patients and upgrade their quality of life. However, the unexplained gait recognition model decreases the prediction accuracy of the exoskeleton system. The existing explainable models are seldom used in the domain of gait recognition due to their high complexity and large computation. To strengthen the transparency of the model, SHapley Additive exPlanations (SHAP) is applied to gait recognition for the first time in this paper, and an interpretable model framework that can be applied to any lower limb exoskeleton is proposed. Compared with the existing methods, SHAP has a more solid theoretical basis and more efficient calculation methods. The proposed framework can find the relationship between input features and gait prediction, to identify the optimal sensor combination. Additionally, The structure of the gait recognition model can be optimized by adjusting the feature attention of the model with the feature crossover method, and the accuracy of the model can be upgraded by more than 7.12% on average.
The time-limited order is a new type of real-time delivery service that platforms need to complete the order delivery with different time granularity (e.g. one day, two days, or three days). Predicting the number of time-limited orders plays an important role for real-time order delivery allocation and anomaly detection in logistics IoT scenarios. However, the impact between orders of different time granularity is complex and the contribution of the static order features is unknown. Previous order predicting methods are not suitable for time-limited orders because they do not fully consider the dependencies among orders with different time granularity. In this paper, we propose a spatial-temporal framework based on stacked long short-term memory networks (LSTM) and deep cross network (DCN) to take the dependencies of multi-scale temporal features into account and fuse cross-domain static features effectively. In addition, we utilize a multi-head attention mechanism to model the heterogeneous strengths of different dependencies. We evaluate our model on a real-world dataset with about 400,000 orders from one of the largest logistics companies in China. The evaluation results show that the Mean Absolute Error and R2 score of our method achieve 9.407 and 0.948, outperforming state-of-the-art solutions.
Mobile social networks facilitate people’s communication and exchange. However, the dissemination of rumor in mobile social networks is not conducive to building a harmonious network environment. Studying the dissemination mechanism of rumor in mobile social networks can reduce the harm caused by rumor to the network environment. Hence, we propose a new model of rumor dissemination in mobile social networks considering the difference of user’s reaction when receiving information and the difference of user’s behavior when processing information. Then, we make a stability analysis of the model. Finally, we conduct simulation experiments on the dissemination process of rumor in mobile social networks. The simulation results show that our proposed rumor dissemination model can effectively predict the dissemination trend of rumor in mobile social networks. Therefore, our work will contribute to the study of the dissemination mechanism of rumor in mobile social networks.
