Abstract:

Abstract:

Convolutional Neural Networks (CNNs) as a quintessential representation of deep learning, are the most commonly used neural networks in tasks such as computer vision. However, convolution operations typically account for over 90% of the runtime in CNNs, becoming a bottleneck for performance. Additionally, due to the complexity of current hardware and the diversity of workloads, specific optimizations in previous work often lack performance portability. To address this problem, the author introduces BlazerML, an open-source convolution computation library based on auto-generated code templates from TVM, capable of automatically generating high-performance convolution implementations for any input shape. BlazerML is implemented based on the Winograd algorithm, known for its high performance in fast convolution algorithms. Experimental results demonstrate that BlazerML significantly outperforms current state-of-the-art open-source libraries. On x86 CPUs, running common deep learning network forward inferences, it is faster by 1.18—2.47 times, 1.18—2.27 times, and 1.01—1.66 times compared to OnnxRuntime, MNN, and the TVM community version, respectively. On ARM CPUs, for single-layer inference of common deep learning networks, it surpasses ACL and FastConv by 1.26—6.11 times and 1.04—4.28 times, respectively.

Abstract:

Existing methods for video traffic identification are mainly targeted at specific platforms and mostly require capturing full flows, which makes them unsuitable for high-speed networks management. This paper proposes a method for video traffic identification from multi-platforms in the sampled high-speed traffic.This paper analyze multiple video platform transmission protocols, extract features based on their common characteristics to construct a composite feature space, and further process these features to eliminate the effect of sampling on feature stability. Then, feature vectors are extracted and a classification model is trained. In the experiments, high-speed networks traffic with a bandwidth of 10 Gbps and a sampling rate of 1:32 was used. The results showed that the proposed method can quickly identify video traffic from multi-platforms with a precision of over 98%.

Abstract:

In response to the current power plant switch detection methods that are unable to cope with realworld open-set environments and the low accuracy in recognizing rare categories, the target recognition problem is transformed into a similarity measurement issue, and a new algorithm is proposed. The new algorithm is based on the triplet network of deep metric learning, using a ResNet-18 with an added SE Block to extract features, and enhances learning effects by cross-batch mining. To evaluate the performance of the algorithm, a dataset with 3 300 switch images was created. The algorithm was tested on the self-built dataset for closedset testing, open-set testing, and few-shot testing. The experimental results show that the algorithm demonstrates excellent discrimination ability in the closed-set state. It can not only accurately identify the categories in the training set but also effectively distinguish states that were not encountered during training and those with lower occurrence frequencies. This capability indicates that the algorithm is not only suitable for real-world open-set environments but also significantly improves the recognition accuracy for small-sample data.

Abstract:

In the field of encrypted mobile application traffic classification, traditional methods classify traffic based on the characteristics of bidirectional traffic. However, in actual scenarios, asymmetric routing will cause remote network administrators to only obtain unidirectional traffic, which will reduce the accuracy of traditional methods. Therefore, an encrypted mobile application traffic classification method using only one-way traffic characteristics is designed. Since downlink traffic contains more information than uplink traffic, the payload of downlink traffic is chosen for analysis. Due to the temporal and spatial correlation of mobile application traffic, a bidirectional long short-term memory network is proposed to capture the temporal correlation of data streams, a convolutional neural network is used to learn the spatial correlation of features, and an attention layer is introduced to focus on important features to further improve the recognition accuracy. Compared with the previous methods, this method has a wider range of use, can be applied to both unidirectional and bidirectional traffic scenarios, and uses fewer features to obtain higher accuracy.

Abstract:

In multi-party conversations, identifying the reply-to relation between messages is an important task in the dialogue domain. Existing efforts have not addressed the following two issues related to data distribution: shorter messages tend to appear more frequently, while shorter texts contain less semantic information, which limits the learning ability of the model; the number of positive samples with reply-to relation is often much less than the number of negative samples, leading to data skewness issue during training phase and reducing the model’s performance in processing positive samples. Aiming at the two issues, this paper proposes an improved model based on a pre-trained language model, which firstly mitigates the short text-related issue through dynamic inquiry window modeling; and then copes with the positive sample-related issue through position-driven positive sample weight optimization. The paper is compared with previous research, and the experimental results show that this paper’s work improves the recall metric by an average of 15.7% compared to the baseline model based on the pre-trained language model. In addition, this paper constructs a new dataset collected from the Telegram platform, which can provide data support for subsequent related studies.

Abstract:

Metaphor has the purpose of inspiring understanding and persuading others. Currently, metaphor presents the trend of multimodal integration of text, images, and videos. Therefore, identifying the metaphorical semantics contained in multimodal contents has research value for Internet content security. Due to the lack of multimodal metaphor datasets, it is difficult to establish research models. Therefore, current scholars pay more attention to text-based metaphor detection. To overcome this shortcoming, the paper first generates a new multimodal metaphor dataset from the perspectives of image-text, metaphor appearance, emotion expression, and author intention. Then, Kappa scores were used to assess the consistency among the annotators of the dataset. Finally, a multimodal metaphor detection model is constructed to verify the quality and value of the multimodal data set by combining image attribute features, image entity features, and text features with the help of a pre-training model and attention mechanism. The experimental results show that the metaphor dataset with emotion and intention can improve the effectiveness of metaphor model detection, and confirm that the interrelationship of multimodal information is helpful for understanding metaphor.

Abstract:

With the rapid development of Internet technology, network security issues have become increasingly prominent. Among these, the identification and classification of encrypted traffic have emerged as significant research directions. This paper provides a comprehensive review of current machine learning-based techniques for encrypted traffic classification. First, it briefly introduces common encryption protocols and their characteristics from a layered perspective. Then, it provides an overview of the datasets and evaluation metrics used in this field. Furthermore, a discussion on encrypted traffic analysis methods based on traditional machine learning and deep learning is conducted, with a detailed analysis of key techniques such as feature engineering and classifier models. Finally, it summarizes the challenges currently faced in this field, including the lack of interpretability and the risk of adversarial examples, and looks ahead to future research directions aimed at enhancing interpretability, automating feature extraction, and automating optimizing model structures.

Abstract:

The unique financial characteristics make smart contract, which are widely nested in various blockchain platforms, one of the most successful applications of blockchain technology. Due to the high economic value of carrying large amounts of assets and virtual currencies, smart contracts are constantly subject to various security attacks. In addition, features such as anonymity and automatic execution make smart contracts used in a variety of illegal transactions and malicious applications. Based on this, the paper firstly introduces the operation mechanism and principle of smart contracts in blockchain-related technologies, discusses the application scenarios of smart contract technology and the potential security vulnerabilities and security problems that may be triggered by the development of smart contract technology, and then, based on the summary of the existing research work, discusses the challenges faced by the field of smart contract vulnerability detection, and looks forward to the future research direction of smart contracts in combination with deep learning technology.

Abstract:

Chiplet-based multi-chip integration designs provide a flexible and scalable solution that surpasses traditional SoC (System on Chip) monolithic integration. However, inter-chiplet communication has become a significant bottleneck affecting overall system performance. The Network on Interposer (NoI) plays a pivotal role in multi-chip systems, directly influencing both performance and development costs. In this paper, we review NoI communication topologies for heterogeneous chiplets. We thoroughly explore the importance of current inter-chiplet communication architectures and discuss their design and implementation methods. This paper covers the entire communication process, spanning from protocol and interface layers to the application layer, classifying interconnect topologies based on their structural configurations and providing in-depth analyses and cross-comparisons for each category. Furthermore, we investigate future directions in NoI communication technologies, identifying technical challenges and potential solutions. We also propose advanced evaluation methods and modeling techniques for reusable interposer layers and topologies. This review aims to provide researchers with a thorough understanding of the current landscape and future trends in NoI technology, emphasizing its crucial role in advancing next-generation semiconductor devices across a wide spectrum of applications.

Abstract:

BACKGROUND:For cone beam computed tomography (CBCT), there has long been a wish to modulate the intensity and distribution of the X-rays to accommodate the patient""s anatomy as the gantry rotates from one projection to another. By doing so, both image artifacts and radiation dose would be reduced. However, the current beam modulation setups, such as dynamic bowtie filters, may be too complex for practical use in clinical applications. OBJECTIVE: This study aimed to investigate a simplified dynamic beam filtration strategy for CBCT imaging to reduce image artifacts and radiation dose. METHODS: In this study, the beam filtration was designed to vary dynamically as the gantry of the CBCT rotates around the object. Specifically, two distinct parts were integrated together: the sheet filter part and the bowtie filter part. The dynamic beam filtration setup has two working schemes, one is a combination of dynamic sheet filter and dynamic bowtie filter, denoted as dynamic filter-dynamic bowtie (DFDB); the other is a combination of dynamic sheet filter and static bowtie filter, denoted as dynamic filter-static bowtie (DFSB). Numerical imaging experiments were performed with respect to three human body parts: shoulder, chest, and knee. In addition, the Monte Carlo simulation platform MC-GPU was used to generate the dose distribution maps. RESULTS: Results showed that the use of the proposed DFDB and DFSB beam filtration schemes can significantly reduce the image artifacts and thus improve the CBCT image quality. Depending on the scanned object, the total radiation dose may be reduced by 30%. CONCLUSIONS: The proposed simple dynamic beam filtration strategy, especially the DFSB approach, might be helpful in the future to improve the CBCT image quality with reduced image artifacts and radiation dose.

Abstract:

The accurate reconstruction of industrial component edges is essential and crucial for visual positioning and quality inspection. To address the issue of difficulty in accurately reconstructing point clouds at the edges of industrial components, a three-dimensional reconstruction algorithm based on point cloud projection is proposed. First, the three-dimensional point cloud of the components is obtained by scanning using a binocular structured light method, edge points in the scanned point cloud are extracted. Then the image edge points are extracted from the binocular images. Subsequently, the point cloud edge points are projected onto the binocular images, the nearest image edge points are searched around each projected point to obtain corresponding binocular edge points. Finally, accurate three-dimensional edge point clouds are reconstructed using stereo vision methods. Experimental results demonstrate that compared to other current methods, this approach can effectively address the issue of false edges caused by interference such as reflection and surface scratches, the reconstructed edge point cloud using this method has high accuracy with reconstruction error less than 0.15 mm and can be applied in industrial scenarios such as bin picking, online quality inspection.

Abstract:

Ketamine, an N-methyl-D-aspartate receptor (NMDAR) antagonist, is clinically utilized for sedation, anesthesia, and the treatment of refractory depression. However, its addictive properties restrict its clinical application. A dose of 0.5 mg/kg is commonly used as an antidepressant in clinical settings, while 15 mg/kg represents the dose typically abused. The effects of varying doses of ketamine on brain network activation remain unclear. In this experiment, two representative doses of ketamine, 0.5 mg/kg and 15 mg/kg, were administered via intraperitoneal injection for 7 consecutive days. Brain network activation was assessed by examining the expression of the immediate early gene protein (cFos). Results indicated that, compared to the saline control group, 0.5 mg/kg ketamine significantly increased the number of cFos-positive cells in the medial prefrontal cortex, intermediate lateral septal nucleus, and periaqueductal gray matter. Conversely, 15 mg/kg ketamine significantly increased cFos expression in the nucleus accumbens, lateral habenula, hippocampal CA3 region, amygdala, and ventral tegmental area. These findings suggest that ketamine"s activation of brain networks is dose-dependent, with different doses activating distinct brain regions. This study lays a foundation for investigating the neuropharmacological effects of different ketamine doses and provides a reference for identifying brain regions associated with its antidepressant and addictive properties.

Abstract:

The usage of mmWave radar for non-contact vital signs monitoring has shown great potentials in the medical and healthcare fields, which enables continuous and imperceptible identity verification. Due to the complex impact of various factors on heart movement, the FMCW mmWave radar can better monitor and capture heart data during sleep, and the obtained heart data can be recognized and classified based on the uniqueness of personal heart movement characteristics. In this study, we propose a deep convolution neural network for identification recognition from one-dimensional time series data of the heart radar signal. The results were compared with 3 SOTA methods, i.e. LSTM, InceptionTime and LSTformer. All the models achieved classification accuracies about 90% on an experimentally acquired heart signal data set in sleep posture. The InceptionTime model has the highest accuracy, but it takes the longest time. The LSTM and LSTformer models have the lower accuracy but the shorter calculation time. The proposed CNN model can obtain similar accuracy but better efficiency in comparison with InceptionTime model.

Abstract:

Against the backdrop of Moore''s Law approaching its limit and the difficulty and surging cost of next-generation integrated circuit technologies, advanced substrate technology is an important carrier to support huge I/O enhancement as well as system integration in the field of advanced packaging, and is one of the core components in the post-Moore era. Currently, semi-additive process based on build-up film (BF) is one of the main ways to realize fine-pitch multilayer packaging substrates. In view of the increasingly prominent problem of signal integrity when electronic equipment operates in high-frequency and high-speed scenes, this paper deeply discusses the influence of physical property of BF materials and structural characteristics on signal transmission loss. Based on typical substrate structures such as microstrip lines and vias, the relationship between BF material parameters and signal transmission performance is studied by electrical simulation analysis system. It is found that in microstrip structure, the signal transmission loss increases with the increase of frequency, and this loss is closely related to the dielectric loss factor of BF material. However, in the via structure, the dielectric constant of BF material has a significant influence on the equivalent capacitance and impedance extreme value, and then affects the impedance mismatch. Although the characteristics of BF material have some influence on impedance mismatch, the design of via structure itself is still the main factor affecting impedance matching. In addition, the conductor loss caused by conductor skin effect increases with the increase of copper foil roughness at high frequency, which provides an important reference for the quality control of copper foil in the manufacturing process of packaging substrate. This study reveals the influence mechanism of BF material and structural characteristics on signal transmission loss, which provides a theoretical basis for the design and optimization of BF material with improved physical properties for packaging substrate.

Abstract:

Data provenance technology is capable of recording and tracking the origins of sensitive documents to prevent their leakage. Traditional network path tracing methods are ineffective in tracking offline documents, and key tracing for encrypted files does not ensure reliable provenance for shared files. Existing techniques such as annotation, reverse querying, and data watermarking often require user involvement and are implemented at the application layer, resulting in inadequate security, lack of transparency and flexibility, and insufficient overall system scalability. This paper introduces an innovative script-based dynamic fingerprint provenance architecture that utilizes modifications to the Linux kernel to achieve foundational provenance, enhancing the security and transparency of document tracing. The fingerprint tracking algorithm is implemented through user scripts, improving the flexibility and effectiveness of document provenance. Additionally, the fingerprint-driven algorithm is designed to meet the demands of multi-load sharing, ensuring efficient and scalable document sharing. Upon verification, this architecture has a minimal impact on the operating system and exhibits excellent scalability. In scenarios involving single or multiple load sharing, the fingerprint-driven algorithm demonstrates transparency, real-time performance, and efficiency.

Abstract:

In this work, a new paradigm of visual language modeling is introduced in ophthalmic image disease recognition. And a multi-disease recognition algorithm based on a pre-trained model of contrasting language images is proposed. First, a new multi-labeled fundus image dataset MDFCD8 containing 8 categories is constructed based on several publicly available fundus image datasets. Then, the generative artificial intelligence GPT-4 is utilized to generate expert knowledge describing the fine-grained pathological features of fundus images, which solves the problem of the lack of text labels in fundus image datasets. The experimental results showed that, the proposed method outperforms the traditional convolutional neural network and Transformer network by 4.8% and 3.2%, respectively. This study also conducted ablation experiments on each module to validate the effectiveness of the method, and also demonstrated the potential of visual language modeling in ophthalmic disease research.

Abstract:

Adenoid hypertrophy (AH) is a key contributor to pediatric obstructive sleep apnea syndrome (OSAS). Physicians rely on nasopharyngeal endoscopy to identify AH and the obstruction of adenoid to the airway. However, due to the limitations of 2D endoscope images, physicians have to infer the 3D structure of the adenoid region, which heavily relies on their expertise and the angle at which the adenoids are observed. The adenoid area is composed of mucosal tissue covered by nasal secretions, which may cause strong reflectivity, sparse features, smooth scenes, and blurred images. Based on these unique characteristics of the adenoids, this paper introduces a multi-view stereo algorithm based on endoscopic image sequences of the adenoid nasopharyngeal cavity. The algorithm employs multi-view stereo to first estimate a depth map corresponding to the images. Subsequently, it utilizes mesh surfaces to fit the rough depth information in the depth space, resulting in smooth and refined depth maps. This leads to a dense and precise reconstruction of the adenoid region. Both synthetic and real experimental results demonstrate that the algorithm can achieve accurate, dense, and smooth reconstruction of the adenoid area, surpassing the existing reconstruction algorithms significantly.

Abstract:

This paper presents a design of single-joint biomimetic robotic fish with compact structure and high swimming efficiency. It allows for convenient disassembly and assembly of pectoral fins, pelvic fins, and caudal fins. The influence of pectoral and pelvic fins on swimming performance was studied via underwater experiments. In the prototype swimming tests, a "binocular vision system" for tracking and recording the motion of the robotic fish was constructed using a high-speed camera and a flat mirror. It enabled tracking and recording of the three-dimensional position information of two marked points on the foremost end of the fish head and above its center of mass. This system provided data support for the quantitative analysis of the swimming performance, posture changes, and head stability of the robotic fish. The results indicated that the robotic fish have good performance in linear propulsion and turning. In the stability experiments, the head stability of the robotic fish equipped with pectoral fins and pelvic fins is better during low-frequency swimming. But no advantage is shown during high-frequency swimming, which is consistent with the phenomenon of various fins of fish in the natural environment being close to the body during high-frequency swimming except for the caudal fin.

Abstract:

LOGO is a new distributed computing framework using a Non-MapReduce computing paradigm. Under the LOGO framework, big data distributed computing is completed in two steps. The LO operation runs a serial algorithm in a number of nodes or virtual machines to process independently the random sample data blocks, generating local results. The GO operation uploads all local results to the master node and integrate them to obtain the approximate result of the big data set. The LOGO computing framework eliminates data communication between nodes during iterations of the algorithm, greatly improving computing efficiency, reducing memory requirements, and enhancing data scalability. This article proposes a new distributed machine learning algorithm library RSP-LOGOML under the LOGO computing framework. A new distributed computing is divided into two parts: the serial algorithm executed by the LO operation and the ensemble algorithm executed in the GO operation. The LO operation can directly execute existing serial machine learning algorithms without the need to rewrite them according to MapReduce. The GO operation executes ensemble algorithms of different kinds depending on the ensemble tasks. In this article, the principle of LOGO distributed computing is introduced first, followed by the algorithm library structure, the method for packaging existing serial algorithms and the ensemble strategy. Finally, implementation in Spark, App development, and the results of performance tests for various algorithms are demonstrated.

Abstract:

The ability to mimic the microenvironment of the human body through fabrication of scaffolds itself a great achievement in the biomedical field. However, the search for the ideal scaffold is still in its infant stage and there are significant challenges to overcome. In the modern era, scientific communities are more attracted to natural substances due to their excess biological ability, low cost, biodegradability, and lesser toxic than synthetic lab made products. Chitosan is a well-known polysaccharide that has recently grabbed high amount of attention for its biological activities, especially in 3D bone tissue engineering (BTE). Chitosan greatly matches with the native tissues and thus stands out as a popular candidate for bioprinting. This review focuses on the potential of chitosan based scaffolds advancement and the drawbacks in bone treatment. Chitosan-based nanocomposites have exhibited strong mechanical strength, water-trapping ability, cellular interaction, and biodegradability characteristics. Chitosan derivatives have also encouraged and provided different routes of treatment and enhanced biological activities. 3D tailored bioprinting have opened new doors to design and manufacture scaffolds of biological, mechanical, and topographical properties.

Abstract:

In today''s digital age, Nginx has emerged as the most prevalent web application server on Linux systems, securing the top position in market share. Given its critical role in ensuring the quality of service for users, optimizing the performance of Nginx servers is important. Despite the widespread deployment of Nginx servers across the two main hardware architectures, X86 and ARM, a comparative analysis of performance tuning on these architectures remains unexplored. This study aims to bridge this gap by employing automatic system parameter tuning on Nginx across these architectures, revealing the significant difference. When handling dynamic requests, the optimized performance of Nginx on X86 architecture significantly outperforms that of the ARM architecture. As a result, the optimized performance of Nginx on X86 architecture achieves a P99 latency of 515 milliseconds, which is performance improvement of 287% than that of the ARM architecture. Conversely, when processing static requests, the ARM architecture demonstrates superior performance, with a P99 latency of 220 milliseconds, resulting in a performance increase of 60% than that of X86 architecture. These findings highlight the distinct advantages of X86 and ARM architectures in handling different types of loads. It shows the significant impact of hardware architecture on optimizing Nginx’s performance. Therefore, to optimize the performance of Nginx web server, system administrators must consider the performance differences between static and dynamic requests of Nginx and the unique iterative efficiency over different hardware architectures.

Abstract:

Artificial intelligence interpretability refers to the ability of people to understand and interpret the decision-making process of machine learning models. Research in this field aims to improve the transparency of machine learning algorithms, making their decisions more trustworthy and explainable. Interpretability is crucial in artificial intelligence systems, especially in sensitive and critical decision-making domains such as healthcare, finance, and law. By providing interpretability, people can better understand the reasoning behind the model''s decisions, ensuring that they are fair, robust, and ethical. In the continuously evolving field of artificial intelligence, enhancing the interpretability of models is a key step towards achieving trustworthy and sustainable AI. The article outlines the development history of interpretable artificial intelligence and the technical characteristics of various interpretability methods, with a particular focus on interpretability in the medical field. It provides a more in-depth discussion of the limitations of current methods on medical imaging datasets and proposes possible future directions for exploration.

Abstract:

As food plays an important role in people''s daily lives, a food map showing the geographical distribution of restaurants in a city is of great social value. Social media has covered every aspect of people''s lives; therefore, social media data provides a wealth of data to support automatic cartography. This work proposes an automatic generation method for urban food maps driven by social media data, integrating machine learning and cartographic algorithms to realize the intelligent generation of stylized urban food maps. A visualization system of urban food maps has been developed, which is applied to four cities, Wuhan, Guangzhou, Chongqing, and Chengdu, for case studies. The results show the effectiveness and good visual expressiveness of our method in presenting urban cuisine for cities.

Abstract:

Action recognition in the dark is a challenging task in practice because it is difficult to learn robust action representations from low light environments. Furthermore, there is a domain gap between dark scenes and the data used by traditional pretrained models, which results in suboptimal results with the traditional pretrain-finetune approach, and pretraining from scratch is costly. To address this issue, a domain-adaptive pretraining method is proposed to improve action recognition performance in the dark environments. The method integrates an external vision enhancement model for de-darkening to introduce critical knowledge for dark scene processing. It also employs a cross-domain self-distillation framework to reduce the domain gap of visual representations between illuminated and dark scenes. Through extensive experiments in various dark environment action recognition settings, the proposed approach can achieve a Top-1 accuracy of 97.19% on the dark dataset of fully supervised action recognition. In the source-free domain adaptation on the Daily-DA dataset, the accuracy can be improved to 49.11%. In the multi-source domain adaptation scenario on the Daily-DA dataset, the Top-1 accuracy can reach 54.63%.

Abstract:

Effectively transferring knowledge from pre-trained models to downstream video understanding tasks is an important topic in computer vision research. Knowledge transfer becomes more challenging in open world due to poor data conditions. Many recent multimodal pre-training models are inspired by natural language processing and perform transfer learning by designing prompt learning. In this paper, we propose an LLM-powered domain context-assisted open-world action recognition method that leverages the open-world understanding capabilities of large language models. Our approach aligns visual representation with multi-level descriptions of human actions for robust classification, by enriching action labels with contextual knowledge in large language model. In the experiments of open-world action recognition with fully supervised setting, we obtain a Top-1 accuracy of 71.86% on the ARID dataset, and an mAP of 80.93% on the Tiny-VARIT dataset. More important, our method can achieve Top-1 accuracy of 48.63% in source-free video domain adaptation and 54.36% in multi-source video domain adaptation.

Abstract:

Aiming at the performance problems of photodetectors in optical communication, remote sensing and infrared thermal imaging, the research progress of photodetectors in the near-infrared band at home and abroad was discussed. Compared with traditional compound semiconductor materials, new materials such as silicon-based, graphene, tellurium compounds, transition metal dihalogenated compounds and perovskites have unique structures and properties, and are important materials for the preparation of low-power and high-performance photodetectors. This paper mainly expounds the research progress of silicon-based near-infrared photodetectors based on PN and PiN heterojunction structures, introduces the research progress of near-infrared photodetectors based on two-dimensional materials (graphene, tellurium compounds, transition metal dihalogenated compounds) and perovskite materials, and analyzes and compares the performance parameters of related near-infrared photodetectors, which provides ideas for the subsequent research of high-performance near-infrared photodetectors.

Abstract:

With the regulations of wearing helmets while driving the electric bicycle, it is urgent to develop a detection algorithm that can accurately detect whether the drivers are wearing helmets. This paper introduces a novel method to detect the helmets based on the YOLO framework. The branch absorption module is proposed to improve the residual backbone network, then the feature fusion is improved through the channel recombination. Finally, the designed structural fusion pruning is applied to further compress the hyper-parameters of the model. The experimental results showed that, the proposed algorithm has higher accuracy and faster speed. Performance of small targets detection also can be improved, with the average accuracy of multiple classification up to 88.8% and detection speed of 29.5fps, which can meet the demand of video surveillance in real applications.

Abstract:

Nowadays, with the rapid development of the Internet, more and more new applications appear gradually, the scale of network expand constantly, and the architecture of network is more and more complicated. As one of the basic technologies for enhancing network controllability, traffic classification can not only provide better QoS for ISPs, but also supervise and manage network effectively, which can ensure the security of the Internet. In this paper we review the research methods and achievements in the field of traffic classification, compare these traditional methods, and point out their advantages and disadvantages. On the other hand, for the real challenges of real-time classification of high-speed network environment, encrypted traffic classification, fine-grained traffic classification, and dynamically changed protocols classification, we describe and analyze the related research progress. Finally, we look ahead the future research direction.

Abstract:

Repetitive sequences are prevalent in genomes. A large number of experiments have confirmed that they play an important role in biological evolution. At present, the discovery and detection of the repetitive sequences have been becoming a hot topic of genomics. This paper summarizes the research progress in this regard, and briefly analyses the associated tools. Finally, the development of repetitive sequences in future is prospected.

Abstract:

The human gut is densely populated by the gut microbiota. There are accumulating evidences indicating that the gut microbiota plays a significant role in the function of the body, which including the metabolism and energy absorption, the development in the function of gastrointestinal, the modulation of immune system and so on. Many chronic diseases, such as obesity, obesity-associated inflammation, inflammatory bowel disease and depression, are related to gut microbiota dysbiosis. The research of the interaction between intestinal bacteria and human body is instructive to the prevention or treatment of many chronic diseases and maintaining health.

Abstract:

The technology of the robot represents a national high-tech level and the degree of automation. It is helpful to develop the industry of service robots in China if we know the current situation and development trend of service robots research clearly. Recently, robotic cleaners and educational robots have been in great demand. Entertainment robots and surveillance robots are developed rapidly and the market expands quickly. Medical robots begin to enter the modern life and have played an important role in the modern surgery. To satisfy the great market and shorten the distance between China and developed countries, it is necessary to capture the development trend of the technology of service robots. R&D on service robots should focus on the integrated technologies on intelligence, modularization and network.

Abstract:

The accurate contour delineation of the target and organs at risk (OAR) is essential in treatment planning for image guided radiation therapy. In clinical applications, the contour delineation is often done manually by clinicians, which may be accurate, but time-consuming and tedious for users. Although a lot of automatic contour delineation approaches have been proposed, few of them can fulfill the necessities of applications in terms of accuracy and efficiency. In this work, a novel approach of target delineation was proposed. Target delineation of OARs was achieved by using snake model and multiscale curve editing to obtain promising results. It allows users to quickly improve contours by a simple mouse click. Experimental results demonstrate the effectiveness of the proposed method for clinical target delineations.

Abstract:

Gene is the genetic material basis. All life phenomena, like disease and death, are related to Gene. Gene sequencing is a way to read life. With the development of new generation high-throughput sequencing technology, TB or more sequence data will be generated daily. It’s more difficult to interpret these big and complex data than to acquire them. Sequence data interpretation is a critical step in current biological research and has great practical significance. It’s a great challenge for current computer systems and computing models to store, process and analysis massive high throughput sequence data. With survey, especially from BGI (Beijing Genome Institute), the current status, problems and measures taken to process high throughput sequence data will be discussed. However, the challenge is too big to be solved unless more people in different fields work together in depth for a long term.

Abstract:

This paper gives a comprehensive introduction to the status of current machine translation research and technology, and analyzes the key problems to be resolved. Finally our idea of the future trends and prospects of machine translation are put forward.

Abstract:

As an advanced biosensing strategy, the electrochemical biosensor is made up of the active sensing biomaterial and the electrochemical signal transducer, and is being widely applied in the fields of clinical medicine, drug and food analysis as well as the environmental monitoring. The electrochemical biosensor has the advantages of excellent specificity, high sensitivity and simplicity. This review focused on the fundamental principle of electrochemical biosensor, its classification and the biomedical applications. The prospect of the electrochemical biosensors was outlined as well.

Abstract:

With the increasing concerns of global warming and resource constraints, electric vehicles (EVs) have made great progress during the past decade. The electric driving system of EVs has dinstinct advantages, such as quick response, easy measurement , and precise control of motor torque, available flexible driving architecture, and regenerative braking, etc. Such advantages can be used to improve the performance of vehicle dynamic control. This paper presents the recent research efforts on electric vehicle dynamic control in terms of parameters estimation and dyanmic control scheme and methodology, especially focusing on the tire-road friction estimaion , novel traction control methods. The lateral dynamic control including the electrical differential control, direct yaw moment control, and the integratin chassis cotrol is proposed. Several prospects for vehicle dynamic control are proposed.

Abstract:

Automatic drive is an important application field of artificial intelligence. In this paper, a novel training strategy for self-driving vehicles was investigated based on the deep reinforcement learning model. The proposed method involves a Q-learning algorithm with filtered experience replay and pre-training with experiences from professional drivers, which accelerates the training process due to reduced exploration spaces. By resampling the input state after clustering, generalization ability of the strategy can be improved due to the individual and independent distribution of the samples. Experimental results show that, in comparison with conventional neural fitted Q-iteration algorithm, the training efficiency and controlling stability can be improved more than 90% and 30% respectively by the proposed approach. Experimental results with more complex testing tracks show that, average travel distance can be improved more than 70% in comparison with the Q-learning algorithm by the proposed method.

Abstract:

Transcranial magnetic stimulation (TMS) is a non-invasive technique that can be used for brain studying and clinical therapy. Firstly, the technology feature and application of the TMS instrument were introduced. Then several TMS coil positioning methods were evaluated and several key problems about TMS coil positioning were discussed. The aim of this study was to propose a new method for TMS coil positioning. The new method combines three aspects of quantitative information including the brain scalp, brain anatomy and brain function and has great advantages and broad application prospects.

Abstract:

Hadoop job schedulers typically use a single resource abstraction and resources are allocated at the level of fixed-size partition of the nodes, called slots. These job schedulers ignore the different demands of jobs and fair allocation of multiple types of resources, leading to poor performance in throughput and average job completion time. This paper studies and implements a Muti-resource Fair Scheduler (MFS) in Hadoop. MFS adopts the idea of Dominant Resource Fairness (DRF). It uses a demand vector to describe demands for resources of a job and allocates resources to the job according to the demand vector. MFS uses resources more efficiently and satisfies multiple jobs with heterogeneous demands for resources. Experiment results show that MFS has higher throughput and shorter average job completion time compared to Hadoop slot-based Fair Scheduler and Capacity Scheduler.

Abstract:

Wireless Capsule Endoscope (WCE) is a very promising tool for the examination of the gastrointestinal (GI) tract. However, there are some problems to be solved for the existed WCE, and one key problem is the accurate localization and tracking of the WCE. Among the possible localization methods, the magnet-based localization technique has its advantages: no need for power, not much space occupation, continuously tracking ability, and no negative effect. In this paper, we present the localization method for the magnet objective inside the WCE based on the magnetic sensor array outside the human body. Through the algorithm and system design we realize real time tracking of 3D position and 2D orientation of the magnet based on the magnetic dipole model. In order to overcome the interference of the human body movement, we propose the multi-magnets’ localization method; also, the 3D positioning and 3D orientation method is proposed, which can be used to make the 3D recovery of the GI tract and the accurate computation of the physiological tissue parameters. The real experiments show that the proposed localization system can run well and obtain the accuracy with 2~3mm for the magnet.

Abstract:

For the advantages of noninvasive, real-time and quantitative detection, ultrasonic transient elastography has important clinical application value. This work investigates the transient elastography in a few ways and aims to design a transient imaging system. The Displacement tracking algorithm based on correlation techniques and the parabolic interpolation algorithm is proposed to improve the accuracy. A novel match filter is designed to convolute with the estimated displacement in the time direction to boost the SNR of the displacement for a better strain image mapping. The convoluted result shows the match filter can significantly improve the strain image quality and help getting more accurate Youngs modulus estimation. The Time Gain Compensation (TGC) circuit is designed to compensate the attenuated power of the ultrasound signal. And a modified polyacrylamide gel based tissue-mimicking phantom is also developed in this paper, both indentation testing and transient elastography are used to characterize the elastic properties of this phantom. The results are almost consistent with each other.

Abstract:

This paper introduces the scientific computing grid, ScGrid, and it’s middleware SCE. ScGrid is built as one virtual supercomputer, integrating computing resource from more than 30 institutes. It provides unified,?easy to use and reliable scientific computing services. SCE is a lightweight grid middleware, which supports global job scheduling and unified data view. It provides multiple user interfaces including command line, grid portal and APIs. At present, ScGrid has been very successfully used in Chinese Academy of Sciences and widely accepted by more than 200 users.

Abstract:

Effectively reducing power line interference is always an important issue in electromyography (EMG) signal recordings and analysis. In this study, four commonly used de-noising methods, including digital notch, LMS based adaptive filter, Kalman filter and S transform, which may be suitable for the reduction of power line interference in real-time EMG recordings, were chosen and their performance in reducing the power line interference from EMG signal recordings were quantitatively analyzed and compared. The pilot results of this study showed that Kalman filter presented the best whole performance in attenuating power line interference from EMG signals and S transform de-noising method illustrated the best performance when the power line interference was severe.

Abstract:

With the rapid increase in numbers and scales of deep web sites on the Internet, search for data or information from deep web sites by submiting queries to and obtaining results from the backend databases has become a major means in information retrieval from the Web. This area has attracted many researchers to devote their efforts on development of technologies to make better use of information in th deep web. One challenge is searching for and integration of data from various databases in deep web. Since deep web is dominated by text data, research and development of technologies for text information retrieval from deep web have a broad application potential. In this paper, we review the state-of-the-art of deep web research in details and propose some future research directions.

Abstract:

Several of the top ranked supercomputers are based on the hybrid architecture consisting of a large number of CPUs and GPUs. High performance has been obtained for problems with special structures, such as FFT-based imaging processing or N-body based particle calculations. However, for the class of problems described by partial differential equations (PDEs) discretized by finite difference (or other mesh based methods such as finite element) methods, obtaining even reasonably good performance on a CPU/GPU cluster is still a challenge. In this paper, we propose and test an hybrid algorithm which matches the architecture of the cluster. The scalability of the approach is implemented by a domain decomposition method, and the GPU performance is realized by using a smoothed aggregation based algebraic multigrid method. Incomplete factorization, which performs beautifully on CPU but poorly on GPU, is completely avoided in the approach. Numerical experiments are carried out by using up to 32 CPU/GPUs for solving PDE problems discretized by FDM with up to 32 millions unknowns.

Abstract:

Foot drop is the inability to voluntarily dorsiflex the ankle during the swing phase of gait and is usually caused by weakness and damages of the peroneal nerve. The consequences of the foot drop include the decreasing of gait quality, the limiting of mobility, the increasing of falling risk, and great increasing of energy expenditure during walking. Firstly biosignal sensors are used in the drop foot stimulator to detect foot movements. Then the surface drop foot stimulator produces a predefined stimulation profile to the peroneal nerve or tibialis anterial to elicit a dorsiflexion of the foot synchronized with the swing phase of gait to lift the foot. This paper reviewed the fundamentals and current researches of drop foot stimulators. Moreover, the development trends of the closed loop drop foot stimulator were also discussed in the paper.

Abstract:

Human action recognition acts as an important role in human machine interaction. This paper proposes a human body recognition method from depth image based on part size and position features. Random forest classifiers are trained with different parameters. Experimental results demonstrate the feasibility of proposed approach. Recognition accuracy is about 91% and the computation time is about 0.96 us per pixel.

Abstract:

Random Forests is an important ensemble learning method and it is widely used in data classification and nonparametric regression. In this paper, we review three main theoretical issues of random forests, i.e., the convergence theorem, the generalization error bound and the out-of-bag estimation. In the end, we present an improved Random Forests algorithm, which uses a feature weighting sampling method to sample a subset of features at each node in growing trees. The new method is suitable to solve classification problems of very high dimensional data.

Abstract:

In order to help the hearing loss children, we obtained hearing loss children’s fallible pronunciation texts and the confusing pronunciation text pairs form a good deal of hearing loss children’s audio pronunciation data. We designed a data-driven 3D talking head articulatory animation system, it was driven by the articulatory movements which were collected from a device called Electro-magnetic articulography (EMA) AG500, the system simulated Chinese articulation realistically. In that way, the hearing loss children can observe the speaker’s lips and tongue’s motions during the speaker pronouncing, which could help the hearing loss children train pronunciation and correct the fallible pronunciations. Finally, a perception test was applied to evaluate the system’s performance. The results showed that the 3D talking head system can animate both internal and external articulatory motions effectively. A modified CM model based synthesis method was used to generate the articulatory movements. The root mean square between the real articulatory movements and synthetic articulatory movements was used to measure the synthesis method, and an average value of RMS is 1.25 mm.

Abstract:

Morphological parameter measurement of Pseudosciaena Crocea plays an important role in its genetic selection and quality improvement. In this paper, an automatic detecting system which can measure the Pseudosciaena Crocea morphological parameters such as weight, length and body width was developed based on the machine vision and weighing sensor technology. The system can automatically detect the external morphology parameters by the machine vision, and get weight parameters through the weighing sensor. The mean errors of dimensional measurement and weighting are 0.28% and 0.74% respectively, which shows that the developed system can completely meet the requirements of morphological parameter measurement for Pseudosciaena Crocea. It is a new effective method to the automatic detection of fish morphology parameters.

Abstract:

In this work the thermal behavior of the LiNi1/3Co1/3Mn1/3O2 cathode material for soft packed lithium-ion power batteries during charging and discharging at different C-rate were conducted using the ARC (accelerating rate calorimeter) to provide an adiabatic environment. The overall heat generated by the lithium-ion battery during use, is partly reversible and partly irreversible, due to entropy change and joule heating, respectively. It indicates that the heating generation of lithium-ion cell is decided by the C-rate of charge and discharge. The heat is smaller at low C-rate of charge and discharge. For example, the heating generation of battery increases 7.16℃ at 0.2C-rate and the entropy change heat is clearly embodied. The joule heating is more remarkable than the entropy change during charging and discharging at high C-rate. For instance, the heating generation of cell increased 25.63℃ at 1C-rate. The heat generation of charge is less than discharge at the same C-rate. The DC inter insistence of cell at the SOC (State of Charge) of 0 to 10% increases suddenly, so the heating generation power will reach its maximum in this period during discharge. It is valuable for the design of heat dissipation in lithium-ion battery thermal management.