With the fast growth of artificial intelligence and big data technologies, AI-based mobile apps are widely used in people' s daily life. However, the quality problem of apps is becoming more and more prominent. Many AI-based mobile apps often demonstrate inconsistent behaviors for the same input data when context conditions are changed. Nevertheless, existing work seldom focuses on performing testing and quality validation for AI-based mobile apps under different context conditions. To automatically test AI-based plant identification mobile apps, this paper introduces TestPlantID, a novel metamorphic testing approach based on test contexts. First, TestPlantID constructs seven test contexts for mimicking contextual factors of plant identification usage scenarios. Next, TestPlantID defines test-context-based metamorphic relations for performing metamorphic testing to detect inconsistent behaviors. Then, TestPlantID generates follow-up images with various test contexts for testing by applying image transformations and photographing real-world plants. Moreover, a case study on three plant identification mobile apps shows that TestPlantID could reveal more than five thousand inconsistent behaviors, and differentiate the capability of detecting inconsistent behaviors with different test contexts.
Authors: Hongjing Guo (College of Computer Science and Technology, Nanjing University of Aeronautics and Astronautics), Chuanqi Tao (College of Computer Science and Technology, Nanjing University of Aeronautics and Astronautics), Zhiqiu Huang (College of Computer Science and Technology, Nanjing University of Aeronautics and Astronautics),
Hide Authors & Abstract

Due to the prevalence and severe consequences of cyberbullying, numerous research works have focused on mining and analyzing social network data to understand cyberbullying behavior and then using the gathered insights to develop accurate classifiers to detect cyberbullying. Some recent works have been proposed to leverage the detection classifiers in a centralized cyberbullying detection system and send notifications to the concerned authority whenever a person is perceived to be victimized. However, there are two concerns that limit the effectiveness of a centralized cyberbullying detection system. First, a centralized detection system gives a uniform severity level of alerts to everyone, even though individual parents might have different tolerance levels when it comes to what constitutes cyberbullying. Second, the volume of data being generated by old and new social media makes it computationally prohibitive for a centralized cyberbullying detection system to be a viable solution. In this work, we propose BullyAlert, an android mobile application for parents that allows the computations to be delegated to the hand-held devices. In addition to that, we incorporate an adaptive classification mechanism to accommodate the dynamic tolerance level of guardians when receiving cyberbullying alerts. Finally, we include a preliminary user analysis of guardians and monitored users using the data collected from BullyAlert usage.
Authors: Rahat Ibn Rafiq (University of Colorado, Boulder), Richard Han (University of Colorado Boulder), Qin Lv (University of Colorado Boulder), Shivakant Mishra (University of Colorado Boulder),
Hide Authors & Abstract

When users manipulate low memory Android devices, they frequently encounter the application problem of loading slowly because of limited amount of memory. In particular, more applications installed, problems will occur more frequently. We deeply observe the low memory management mechanism of the Android system and find the system has some shortcomings, such as memory recovery efficiency, unnecessary memory requests. In this paper, we optimize memory usage by improving recovery efficiency, prioritize the use of less memory, prevent the instantaneous increase in memory usage, and reduce unnecessary memory requests. Experimental results in a real environment show that our methods effectively increase the size of free memory, and reduce the phenomenon of application self-startup and association startup.
Authors: Linlin Xin (Advanced Institute of Information Technology, Peking University), Hongjie Fan (School of Electronics Engineering and Computer Science, Peking University), Zhiyi Ma (School of Electronics Engineering and Computer Science, Peking University),
Hide Authors & Abstract

Network functions virtualization (NFV) emerges as a promising network architecture that separates network functions from proprietary devices. NFV lowers the cost of hardware components and enables fast and flexible deployment of network services. Despite these advantages, NFV introduces new security challenges. Currently, there is little research on a holistic framework to solve these security issues. In this paper, we propose a security service orchestration framework that can construct a cooperative working mechanism for NFV security. We present the demand analysis and describe the system design principles and implementation details. The system’s effectiveness is also shown based on technical review.
Authors: Hu Song (State Grid Jiangsu Electric Power Co., Ltd. Information and Communication Branch), Qianjun Wu (Information System Integration Company, NARI Group Corporation), Yuhang Chen (Information System Integration Company, NARI Group Corporation), Meiya Dong (Taiyuan University of Technology), Rong Wang (China Energy Investment Corporation),
Hide Authors & Abstract

With the rapid development of industrial demands, the Internet of Things triggers enormous interests by industry and academia. By employing IoT technologies, a large number of problems in the industry can be solved by intelligent sensing, wireless communication, and smart software analysis. However, in applying Industrial IoT to improve real-time and immerse user experiences, we found that compared to traditional application scenarios such as tourism, or daily experiences, industrial IoT applications face challenges in scalability, real-time reaction, and immerse user experiences. In this paper, we propose an edge-assisted framework that fits in industrial IoT to solve this fatal problem. We design a multi-pass algorithm that can successfully provide a real sense of immersion without changing the single frame image visual effect in terms of increasing rendering frame rate. From experimental evaluation, it shows that this edge-assisted rendering frame-work can apply to multiple scenarios in Industrial IoT systems.
Authors: Zeng Zeng (State Grid Jiangsu Electric Power Co., Ltd.), Weiwei Miao (State Grid Jiangsu Electric Power Co., Ltd.), Chuanjun Wang (State Grid Jiangsu Electric Power Co., Ltd.), Shihao Li (State Grid Jiangsu Electric Power Co., Ltd.), Yuze Jin (Digital China), Peng Zhou (Digital China), Hongli Zhou (Digital China), Meiya Dong (Taiyuan University of Technology),
Hide Authors & Abstract

At present, the problem of shortage of medical human resources can be solved through mobile medical equipment; the main method to improve the diagnostic performance of mobile medical equipment is to improve the accuracy of the algorithm. Brain tumor classification is to determine the tumor type of patients. The accurate brain tumor classification algorithm can improve the diagnostic performance of mobile medical equipment while assisting doctors in diagnosis. This paper proposes a multi-grade brain classification system using improved CNN framework with extensive data augmentation for differentiating among glioma, meningioma and pituitary tumors, which from three prominent types of brain tumor. First, we locate the tumor and extract the region of interest (ROI). Secondly, to solve the problem of insufficient data samples in the brain tumor classification, we use data augmentation techniques to augment the data samples. Thirdly, VGG-19 and Inception V3 model are improved, and the CNN model is optimized by Adam algorithm. Finally, the improved CNN framework is trained and classified with augmented dataset. Experimental results show that the system proposed in this paper based on data augmentation and improved CNN framework has better classification performance than traditional classifier, and this system can effectively solves the problem of low accuracy caused by insufficient data samples.
Authors: ning xin (School of Software Technology,Zhengzhou University), li zhanbo (Department of Network Management Center,Zhengzhou University), pang haibo (School of Software Technology,Zhengzhou University),
Hide Authors & Abstract

Chronic Obstructive Pulmonary Disease (COPD) is a major health concern for elders today. Chronic cough and wheezing, which occur in the lungs as a result of mucus buildup are the main symptoms of COPD. COPD patients are advised to regularly medicate themselves via an inhaler, which delivers medicine to the lungs to break down mucus and relieve wheezing. Unfortunately, many patients do not use their inhaler devices correctly, resulting in no improvement of COPD symptoms, and worsened health. In this paper, we design machine learning (Support Vector Machine) algorithms operating on Mel-frequency Cepstral Coefficients of cough/breath sounds of patients ({recorded via smartphones before and after inhaler usage}) to detect the effectiveness of inhaler usage. Using a cohort of $55$ clinically diagnosed COPD patients, spread across both genders, we evaluate our system from multiple metrics, including Precision, Recall, Sensitivity and Specificity. Our system achieved accuracies close to 80% in detecting effectiveness of inhaler usage. Our proposed system can aid COPD patients in improved self-care routines, and also reduce the rate of re-hospitalizations caused by exacerbated symptoms.
Authors: Anthony Windmon (Univ. of South Florida), Sriram Chellappan (Univ. of South Florida), Ponrathi Athilingam (Univ. of South Florida),
Hide Authors & Abstract

How can the use of IoT improve safely navigation for kick scooters? The con-text-aware and location-based solution combines an embedded smart bracelet and a mobile phone. The integrated platform provides a ubiquitous service, which gives kick scooters the ability to safely navigate by reducing distractions. A key motivation was to minimize attention to the external disturbances during navigation and maximize attention to the route itself. This was achieved by (1) providing guidance while navigating via synchronized vibro-tactile feedback on a smart bracelet; (2) developing a vibration language on the bracelet; (3) allowing users to produce navigation routes in-advanced on their mobile phone, or con-sume existing routes in real-time. The solution was designed for inclusive and tested in real-life situations on busy roads in the city.
Authors: Yonit Rusho (Shenkar Engineering, Design, Art), Haim Elbaz (Shenkar Engineering, Design, Art), Reut Leib (Shenkar Engineering, Design, Art), Roni Polisanov (Shenkar Engineering, Design, Art),
Hide Authors & Abstract

Cluster analysis of power loads can not only accurately extract the commonalities and characteristics of the loads, but also accurately understand the users' habits and patterns of electricity consumption, so as to optimize the power dispatching and regulate the operation of the entire power grid. Based on the traditional clustering method, this paper proposes a clustering algorithm that can automatically determine the optimal cluster number. Firstly, Fast-DTW algorithm is used as the similarity measuring function to calculate the similar matrix between two time series, and then Spectral Clustering and Affinity Propagation (AP) algorithm are used for clustering. It is combined with Euclidean distance, DTW and Fast-DTW algorithms to evaluate the algorithm effect. By analyzing the actual power data, the results show that the improved external performance evaluation index ARI, AMI and internal performance evaluation index SSE are significantly improved and have better time series similarity and accuracy. Applying the algorithm to thousands of users, several kinds of typical power load patterns can be obtained. For any other load curve, it can be mapped to a standard load by feature extraction. The corresponding prediction model is adopted, which is of great significance to reduce the peak power consumption, adjust the electricity price appropriately and solve the problem of system balance.
Authors: Zhongqin Bi (College of Computer Science and Technology, Shanghai University of Electric Power, China), Yabin Leng (College of Computer Science and Technology, Shanghai University of Electric Power, China), Zhe Liu (State Grid Shanghai Electric Power Research Institute, Shanghai, China), Yongbin Li (Office of Academic Affairs, Shanghai University of Electric Power, China), Prof. Stelios Fuentes (Leicester University, UK),
Hide Authors & Abstract

Combining the advantages of the convolutional neural network CNN and the minimum gated unit MGU, the attention mechanism is merged to propose an attention C_MGU neural network model. The preliminary feature representation of the extracted text is captured by the CNN's convolution layer module. The Attention mechanism and the MGU module are used to enhance and optimize the key information of the preliminary feature representation of the text. The deep feature representation of the generated text is input to the Softmax layer for regression processing . The sentiment classification experiments on the public data sets IMBD and Sentiment140 show that the new model strengthens the understanding of the sentence meaning of the text, can further learn the sequence-related features, and effectively improve the accuracy of sentiment classification.
Authors: Diangang Wang (State Grid Sichuan information & communication company), Lu Xiaopeng Lu (School of Computer Science and Technology, Shanghai University of Electric Power), Yan Gong (State Grid Sichuan information & communication company), Linfeng Chen (School of Information and Electronic Engineering, Zhejiang University of Science and Technology), Lin Huang (State Grid Sichuan information & communication company),
Hide Authors & Abstract

In order to remove the Gaussian noise in the image more effectively, a convolutional auto- encoder image denoising model combined with the perception module is proposed. The model takes the whole image as input and output, uses the concept module to denoise the input noise image, uses the improved concept deconvolution module to restore the denoised image, and improves the denoising ability of the model. At the same time, the batch normalization (BN) layer and the random deactivation layer (Dropout) are introduced into the model to effectively solve the model over fitting problem, and the ReLu function is introduced to avoid the model gradient disappearing and accelerate the network training. The experimental results show that the improved convolution neural network model has higher peak signal-to-noise ratio and structure similarity, better denoising ability, better visual effect and better robustness than the deep convolution neural network model.
Authors: Diangang Wang (State Grid Sichuan information & communication company), Wei Gan (State Grid Sichuan information & communication company), Chenyang Yan (Shanghai University of Electric Power), Kun Huang (State Grid Sichuan information & communication company), Hongyi Wu (Zhejiang University of Science and Technology),
Hide Authors & Abstract