Sleep staging by highly trained specialists is laborious. Although automatic sleep staging with supervised learning methods has been implemented for almost a decade, it requires lots of manually annotated data. Self-supervised learning methods have recently been gaining attention. They can learn representations with unlabeled data, which alleviates the cost of labeling work. However, the problem is that these self-supervised sleep staging methods either require prior knowledge, as with frequency information, or they produce unsatisfactory results. Thus, we propose a deep sequential sleep network (DSSNet), a self-supervised framework that aims to perform multi-view representations based on contrastive learning. It utilizes a single-channel electroencephalogram but achieves competitive performance. We also explore the impact of different contrastive mechanisms on DSSNet performance. The results of the Sleep-EDF dataset prove that the consistency of negative samples is crucial for improving performance. We evaluate DSSNet on Sleep-EDF and ISRUC-Sleep and achieve accuracies of 80.0% and 71.4%.

Many distributed stream computing engines have emerged to handle big data, and they can be deployed in cloud environments consisting of native networks or container networks. Most of the benchmark research on stream computing engines are carried out under the native network, and the research on the impact on container network on stream computing engines is currently inadequate. However, the use of container network will inevitably lead to performance degradation, which is the disadvantage of all virtual networks. In this work, we build Apache Storm and Apache Flink, which are Streaming Computation Engines in container network and native network environments and conduct performance measurements through experiments processing textual data to verify how much performance decreases in container network. Experiments show that the throughput in a container network environment is 1%-5% lower and CPU utilization is 11%-18% lower than in a local network environment.

Users of recommendation systems usually focus on one topic at a time. When finishing reading an item, users may want to access more relevant items related to the last read one as extended reading. However, conventional recommendation systems are hard to provide the continuous extended reading function of these relevant items, since the main recommendation results should be diversified. In this paper, we propose a new task named recommendation suggestion, which aims to (1) predict whether users want extended reading, and (2) provide appropriate relevant items as suggestions. These recommended relevant items are arranged in a relevant box and instantly inserted below the clicked item in the main feed. The challenge of recommendation suggestion on relevant items is that it should further consider semantic relevance and information gain besides CTR-related factors. Moreover, the real-time relevant box insertion may also harm the overall performance when users do not want extended reading. To address these issues, we propose a novel Real-time relevant recommendation suggestion (R3S) framework, which consists of an Item recommender and a Box trigger. We extract features from multiple aspects including feature interaction, semantic similarity and information gain as different experts, and propose a new Multi-critic multi-gate mixture-of-experts (M3oE) strategy to jointly consider different experts with multi-head critics. In experiments, we conduct both offline and online evaluations on a real-world recommendation system with detailed ablation tests. The significant improvements in item/box related metrics verify the effectiveness of R3S. Moreover, we have deployed R3S on WeChat Top Stories, which affects millions of users. The source codes are in https://github.com/modriczhang/R3S.

Atrial Fibrillation (AF) is considered to strongly correlate with stroke. Deep Neural Networks (DNNs) improve the accuracy in real-time atrial fibrillation detection. However, the deployment of DNNs on embedded systems is challenging due to hardware resources. To reduce computation loads, we study the feasibility to eliminate the redundant information in the AF detection task by downsampling. It is compatible with kernel-level optimization, quantization optimization, and model compression methods. A state-of-the-art deep learning model is used to estimate the amount of AF detection information among different sampling rates. This work considers both fixed-length and variable-length time intervals of an Electrocardiograph (ECG) segment. Experiment results demonstrate that model performance can be retained perfectly in AF detection. Ablation study experiments demonstrate the robustness with downsampled signals. Using a large time interval, the AF detection accuracy with 60 Hz signals can be compared to that with 300 Hz signals. Our on-going work includes designing lightweight DNN models with downsampled signals, further exploring the robustness of downsampled signals and model compression.

The high-frequency characteristics of power cables are very important in many application situations, especially the current distributions of power cables in the high frequency band. This paper mainly presents two important integral equation methods, volume integral equation (VIE) and surface integral equation (SIE), which can be applied to arbitrary shape of cross section of cables. This paper also gives a detailed study of the analytical method for the calculation of the current distribution over circular cross section as well as the partial diffusion equation (PDE)-series method for the calculation of current distribution over rectangular cross section. The current distribution shows strong singularity when the frequency is high enough that the skin depth is smaller than the cable's dimension size. An accurate volume integral equation is presented without approximation by using the Hankel function as the integral kernel. At very low frequencies, it reduces to the widely-used quasi-static VIE. The quasi-static VIE is not accurate enough for the calculation of current distribution. Detailed discussion is given for the different current distributions calculated from different methods.

In order to achieve better energy saving effect of the super-capacitor energy storage system (SC-ESS), an on-line optimization control strategy is proposed in this paper. Firstly, the model of urban rail traction power supply system with super-capacitor energy storage system is established, and the working principle of ESS is analyzed. Then the paper analyzes the influence of train braking characteristics and headway on the energy saving effect, and gives the recommended setting range of charge threshold. A real-time optimization control strategy is proposed to adjust the charge and discharge threshold in real time according to the energy saving in a certain period of time, which can make the best energy saving effect of the SC-ESS. The control strategy based on the combination of offline optimization and real-time threshold adjustment, using genetic algorithm in multi-trains operation condition obtains threshold optimal control parameters, and the threshold is adjusted according to the working state of the ESS in actual operation. Simulation and experimental results demonstrate the effectiveness of the proposed control strategy.

The reuse of the regenerated braking energy has huge potential. This paper mainly focuses on the efficient usage of regenerative energy. When one train is braking, if the regenerative energy isn't absorbed by other trains, the traction supply network voltage will increase, it leads to the regenerative braking failure and the waste of regenerative braking energy. Based on the knowing and analysis of online running trains' operation conditions, the other trains' operation curve will be adjusted correspondingly, to achieve the synchronization of acceleration and braking condition. In order to analyze the adjustment principles and adjustment strategies, the model of traction supply network and the constraint conditions of running train are put forward, then simulation in the simple traction power supply system of an actual line are analyzed to verify the correctness and validity of the methods.

Ensemble classification (the concurrent development of K independent classifications) is a common practice in Gibbs classification. In this paper, we describe an adaptation of R. Azencott's (1992) theorem on finite-time annealing, coupled with a limited merging protocol, which produces a final low-energy Gibbs classification of the Virtual Female (VF) data set.

Both formal and informal locally-adaptive cooling schedules have been suggested to improve the convergence rate of Gibbs (and Gibbs-like) classification algorithms. One strategy involves maintaining a global cooling schedule/visiting schedule which is turned on or off (or forcing extremal temperature values) at a site depending on the interiteration behaviour of the classifier. This (0,1)-valued behaviour of the cooling schedule is parameterized relative to the site. We give a preliminary discussion of a method of assigning such parameters based on a multigrid decomposition of the image. The application domain is mammography.

Object-oriented implementation of Signaling System 7 (OOSS7) sets an example of applying object-oriented methods onto network protocol implementations. Unlike many of the implementations where SS7 was integrated into systems in an exclusive and device-dependent way, OOSS7 inherits the advantages of object-oriented programming: reusability, safety and device independence. In this paper, the object-oriented analysis of SS7 is given based on SDL (specification and description language) and MSC (message sequence chart). The data structure and scheduling algorithm are discussed. The compatibility of TUP (telephone user part) and ISUP (ISDN user part) is proved. The experimental system passed the validation test both in the laboratory and in the central office with real operating traffic.

We examine some of the issues surrounding the development of distributed services medical imaging network (DSMIN). The system described is an agent-enabled, dynamic environment which supports the distribution of image data and code within a network cloud of cooperating agencies. The foundations of the DSMIN include code migration and load-balancing using agents and dynamic discovery of services using both the DII/Trader facilities of CORBA as well as more direct client-creation via Java applets. The paper provides an overview of the role of these technologies in the DSMIN.

Usually, data streams communicated between a source and a sink do so under the control of statically-established agencies (such as coder/decoders) established at either end of the pipe. In this paper, we propose the use of Java-based agent technology to impart control activity between source and sink, so that (for example), codecs can adapt to channel conditions, adaptive encryption of sensitive data (and even code) can be overlaid on the channel, and code can be migrated over the channel in addition to (or even in place of) data.

We describe the architecture of a distributed medical imaging system. Designed to support both code and data mobility, the system attempts to exploit available computational resources within a collaborating group to apportion imaging tasks to the platforms best suited to their completion. The system is designed around a mixed Java-RMI (remote method invocation), CORBA and agent architecture. This paper surveys various features of the system.