Multivariate time series analysis is an important problem in data mining because of its widespread applications. With the increase of time series data available for training, implementing deep neural networks in the field of time series analysis is becoming common. Res2Net, a recently proposed backbone, can further improve the state-of-the-art networks as it improves the multi-scale representation ability through connecting different groups of filters. However, Res2Net ignores the correlations of the feature maps and lacks the control on the information interaction process. To address that problem, in this paper, we propose a backbone convolutional neural network based on the thought of gated mechanism and Res2Net, namely Gated Res2Net (GRes2Net), for multivariate time series analysis. The hierarchical residual-like connections are influenced by gates whose values are calculated based on the original feature maps, the previous output feature maps and the next input feature maps thus considering the correlations between the feature maps more effectively. Through the utilization of gated mechanism, the network can control the process of information sending hence can better capture and utilize the both the temporal information and the correlations between the feature maps. We evaluate the GRes2Net on four multivariate time series datasets including two classification datasets and two forecasting datasets. The results demonstrate that GRes2Net have better performances over the state-of-the-art methods thus indicating the superiority.

Identity authentication plays an important role in maintaining social security. With the rapid growth of population, an important issue is put in front of people. How to identify people more accurately and faster? This paper presents a novel behavioral biometric authentication mechanism for smartphones, named PickupAuth. PickupAuth relies on a combined movement, which is the user picks up the phone to his chest and clicks on the touch screen. PickupAuth uses the built-in smartphone sensors to collect the force of the click screen, the acceleration data, the magnetic field data, and the rotation rate. A new proposed feature subset selection algorithm based on the correlation and weight values and Random Forest classification algorithm are used to analyze the sensor data. Experiments on 110 identity users show that the accuracy reaches 98.44% in sitting posture and 97.46% in standing posture. Besides, PickupAuth is more efficient than state of the art considering the training data acquisition time and the training time.

Energy consumption prediction has been poured lots of attention due to its importance in energy planning, management, conservation, etc. This paper proposes a convolutional network called Fully Convolutional Attention Res2Net (FCARN) based on Res2Net for fast and precise energy consumption prediction. Based on the thought of attention mechanism, gate whose value is calculated based on global feature maps are applied in the Res2Net. It assigns weights to the feature maps thus enabling the model to focus on the more important part. We conducted experiments and evaluated our model on appliances energy prediction dataset. As the thought of the gates is similar to Squeeze-and-Excitation Net (SENet), we compare our model with Res2Net, Res2Net combined with SENet and the existing state-of-the-art models. The results demonstrate the superiority and competitiveness of our model. In addition, we provide details of the training process.

Social platforms such as Weibo, Facebook and Twitter have become a part of daily life, where people can exchange information. In this process, people's behaviors often influence each other. Social influence prediction has become one of the hot issues at present. In this paper, NNMLInf social influence prediction model is constructed based on neural network multi-label classification. People's network structure features are taken as the network input, and their behaviors are divided into multiple labels as the network output. Node2vec is adopted to extract network representative features of users. This model combines the network structure with human behaviors, and the prediction results can be more practical. The experiment carried on BlogCatalog, Flickr and Youtube shows that NNMLInf model performs better than traditional approaches such as DT (decision tree), SVM (support vector machine), and better expresses social influence .

Localization has always been an essential application of underwater acoustic sensor networks (UASNs), which plays an important role in routing strategies design, node recycling, and so on. The complex ocean condition, prior infrastructure deployment, and time synchronization among beacons challenge the application of the UASN's localization. To solve these problems, we propose a novel underwater acoustic sensor networks localization algorithm based on the virtual node assistance. The algorithm is classified into two parts based on the current marine environment, including Virtual node Assisted Static (VAS) localization algorithm and Virtual node Assisted Dynamic (VAD) localization algorithm. An auxiliary node, which does not directly participate in the localization, is deployed for virtual node setup, error measurement, and RSSI ranging. The GPS-equipped ship utilizes virtual node and geometry to realize the UASN's localization without complicated deployment procedures and time synchronization. The simulation results show that our proposed algorithm can achieve the properties, including high localization coverage and small localization error as well as low communication overhead in the UASNs.

Configurable software allows users to customize software behaviors through configurations. However, software misconfigurations that lead to the hard-to-diagnose system crash failures could inflict enormous harm to users and should be diagnosed with a high priority. To address this problem, we present a systematic approach (and its tool implementation, called STAD) to diagnosing misconfigurations based on static code analysis. Our approach analyzes the value dependency between variables obtained by exploring the stack trace, generates the value dependency graph (VDG), recommends the root cause of a misconfiguration via the VDG, and utilizes the correlation between configuration options to improve our recommendation results. There are two advantages compared with existing approaches: STAD does not require software crash reproduction, and users do not need to provide configuration options and their option read points (i.e. the statements that access the values of configuration options). We evaluated STAD on 8 misconfigurations from JChord built on Java. STAD can successfully diagnose all misconfigurations with less average number of false positives compared with existing approaches. In addition, STAD runs in less than one minute for each misconfiguration, making debugging more efficient.

Social platforms such as Weibo, Facebook and Twitter have become a part of daily life, where people can exchange information. In this process, people's behaviors often influence each other. Social influence prediction has become one of the hot issues at present. In this paper, NNMLInf social influence prediction model is constructed based on neural network multi-label classification. People's network structure features are taken as the network input, and their behaviors are divided into multiple labels as the network output. Node2vec is adopted to extract network representative features of users. This model combines the network structure with human behaviors, and the prediction results can be more practical. The experiment carried on BlogCatalog, Flickr and Youtube shows that NNMLInf model performs better than traditional approaches such as DT (decision tree), SVM (support vector machine), and better expresses social influence .

Automatically extracting software configuration options not only dramatically reduces the burden of maintaining configuration manuals, but also is an essential part of tackling configuration errors. Existing approaches focus on the key-value configuration model, and depend heavily on the arguments of configuration APIs to infer configuration option names. However, we find that some software adopts the Spring XML configuration model whose configuration options are the elements and the elements. Specially, these elements are nothing to do with the arguments of configuration APIs. To address this problem, we describe the configuration mechanism of the Spring XML configuration model, and present BeanFinder, a static analysis approach to automatically extracting configuration options for software using this model. An empirical evaluation performed on 3 Java open source software shows the effectiveness of BeanFinder.