Search

With the rise of big data, more and more users will launch computing systems to process a large volume of data in various applications. A Scheduling algorithm is crucial to the performance of the processing platforms, especially when they are concurrently executing a batch of jobs. Such jobs usually represent multiple stages. Each stage produces the intermediate data which will be piped to the next stage for further processing. However, the scheduling problem in a big data computing system is different from the traditional multi-stage job scheduling problem as for any two consecutive stages, the later stage usually starts before the former stage is finished to “shuffle” the intermediate data. In this paper, we consider MapReduce/Hadoop as a representative computing system and develop a new strategy named OMO, Optimize MapReduce Overlap with a Good Start (Reduce) and a Good Finish (Map). A MapReduce job contains two consecutive phases: map and reduce. Our general target is to optimize the internal overlap between these two phases. There are two new techniques included in our solution, Lazy start of reduce tasks and Batch finish of map tasks, which aim to approach an effective alignment of the two phases based on the characteristics of the MapReduce process. OMO has been implemented on the Hadoop system with extensive experiments for performance evaluation. The results show that OMO’s performance is superior in terms of total completion time (i.e., makespan) of a batch of jobs.

Point-of-Interest (POI) recommendation is one of the main functions in Location-based Social Networks (LBSNs). At present, some POI recommendation models couldn’t effectively capture users' complex information and their recommendation performances are not good enough. Motivated by this, we propose a kind of POI recommendation model which is based on Attention-based Gated Recurrent Unit Network (Att-GRU). Firstly, we adopt the Gated Recurrent Unit (GRU) network to learn the complex sequential transition patterns from users’ check-in behavior. Secondly, we extract user’s preferences by using the target-guided attention mechanism. Finally, we conduct extensive experiments on a real-world LBSN dataset, and the experimental results demonstrate that our proposed Att-GRU based POI recommendation model has better performance than those of other mainstream ones in some evaluation criteria.

Liver segmentation is a common difficulty and key problem in the field of medical imaging. Aiming at the problem that the existing liver CT images have various shapes and the boundary regions are difficult to segment. So an improved U-Net liver segmentation method is proposed. First, introduce the CSPP module with a dilated convolution and multi-scale feature fusion structure to expand the receptive field while extracting richer spatial information; secondly, connect the CSPP modules in series and combine the residual structure to form the DREG module and join the U-Net network jump connection structure. The high-level and low-level feature information is fused to retain the subtle edge information of the liver; the above method solves the problem of diverse sample data shapes and difficult segmentation of boundary regions. Experimental results on the MICCAI 2017 Liver Tumor Segmentation(LiTS) challenge dataset show that this method obtains a good accuracy rate and has high application value for the clinical auxiliary diagnosis of liver CT images.

Spherical robots have many advantages compared with traditional wheeled and orbital robots. In this paper, the design and implementation of a novel quadrotor-based spherical robot is presented. The robot, named FBall, consists of a spherical outer shell made of Polymethacrylimide (PMI) foam, four variable-pitch propellers, four motors, a controller and a battery. The robot is capable of rolling on water surfaces, rolling on ground surfaces and flying in the air by changing the pitch angle of each propeller and adopting corresponding control algorithms. In rolling mode, two propellers’ pitch angle are set to negative values, and the other two are set to positive values so as to provide a rolling moment. In the flying mode, the pitch angle of all the propellers are set to the same value just as traditional quadrotors, and 4 PID controllers are employed to regulate its roll, pitch, yaw angles, and altitude.The paper focuses primarily on designing a spherical robot based on a variable pitch quadrotor. Modeling, locomotion capabilities analyzing and future works are also presented.

High quality simulation environment can accelerate the development of the control software. Also, it is a good tool for the beginner to study how the control software works. The goal of this paper is aiming at designing a real time HIL experimental platform for quadrotor autopilot software design, validation and optimization. The platform is mainly composed of four blocks: A quadrotor dynamics model, a flight control system, a communication agent and hardware devices for control input and output. The flight dynamics model of a quadrotor build in this paper is to simulate the physics involved with its movement. It receives input signals from the autopilot controller, and returns the quadrotor’s real time pose, position, velocities, and etc. resulting from the input signals to the autopilot controller. A communication agent implements bidirectional communication between the flight dynamics model and the autopilot controller. Hardware devices for input and output is used to change the flight mode, indicate the flight state and test the control law. This paper focuses primarily on developing an experiment platform that works for control software validation. It allows for the safe testing of experimental code and settings, especially for aggressive and agile flight test.

For traditional electromagnetic detection of steel wire rope damage, there are a lot of noise in the collected signals, and a wavelet denoising method based on soft threshold is proposed to apply to denoising. By comparing the signal-to-noise ratio obtained by using different wavelet bases to denoise the detection signal, it was found that db4 had the most ideal denoising effect on the detection signal. In the experiment, the detection signal of different broken wires was denoised. The experimental results show that the wavelet transform soft threshold denoising method is effective, which provides a guarantee for the subsequent feature value extraction and quantitative recognition accuracy.

We propose novel Stacked Spatio-Temporal Graph Convolutional Networks (Stacked-STGCN) for action segmentation, i.e., predicting and localizing a sequence of actions over long videos. We extend the Spatio-Temporal Graph Convolutional Network (STGCN) originally proposed for skeleton-based action recognition to enable nodes with different characteristics (e.g., scene, actor, object, action), feature descriptors with varied lengths, and arbitrary temporal edge connections to account for large graph deformation commonly associated with complex activities. We further introduce the stacked hourglass architecture to STGCN to leverage the advantages of an encoder-decoder design for improved generalization performance and localization accuracy. We explore various descriptors such as frame- level VGG, segment-level I3D, RCNN-based object, etc. as node descriptors to enable action segmentation based on joint inference over comprehensive contextual information. We show results on CAD120 (which provides pre-computed node features and edge weights for fair performance comparison across algorithms) as well as a more complex real- world activity dataset, Charades. Our Stacked-STGCN in general achieves improved performance over the state-of- the-art for both CAD120 and Charades. Moreover, due to its generic design, Stacked-STGCN can be applied to a wider range of applications that require structured inference over long sequences with heterogeneous data types and varied temporal extent.

Doctors often need to convey complex information, such disease states and surgical procedures, to their patients in a simple and straightforward manner. For example, doctors may encounter difficulty in explaining the information needed for surgical consent, such as in the case of prostate cancer, in a way that an elderly patient can understand. The most direct way to help facilitate this communication is through visualization, specifically using augmented reality (AR) tools to visually illustrate to the patients their prostate, medical condition, as well as the necessary surgical procedures. Our goal is to provide a visualization technology tool which makes medically-related communication between doctors and patients easy and efficient.

Deep learning has undergone tremendous advancements in computer vision studies. The training of deep learning neural networks depends on a considerable amount of ground truth datasets. However, labeling ground truth data is a labor-intensive task, particularly for large-volume video analytics applications such as video surveillance and vehicles detection for autonomous driving. This paper presents a rapid and accurate method for associative searching in big image data obtained from security monitoring systems. We developed a semi-automatic moving object annotation method for improving deep learning models. The proposed method comprises three stages, namely automatic foreground object extraction, object annotation in subsequent video frames, and dataset construction using human-in-the-loop quick selection. Furthermore, the proposed method expedites dataset collection and ground truth annotation processes. In contrast to data augmentation and data generative models, the proposed method produces a large amount of real data, which may facilitate training results and avoid adverse effects engendered by artifactual data. We applied the constructed annotation dataset to train a deep learning you-only-look-once (YOLO) model to perform vehicle detection on street intersection surveillance videos. Experimental results demonstrated that the accurate detection performance was improved from a mean average precision (mAP) of 83.99 to 88.03.

It is difficult to apply traditional testing adequacy criteria when measuring the adequacy of convolutional neural network applications. However, only a small number of test cases applied to the CNN model can achieve neuron coverage of almost 100%, overturning the effectiveness of the neuronal coverage criteria. In this paper, we propose a model coverage criterion based on mutation testing for CNN, and applying model coverage criterion to a common CNN image classification models (LeNet-5). we focus on the testing accuracy of model. Experiments show that our method can find the local optimal model and play an important role in improving the testing adequacy of the set of models.

In this paper, a filtering antenna operating under TM 03 patch mode is proposed without any extra feeding or filter circuit. Two radiation nulls at two edges of the operating band are obtained by introducing slots at a radiative edge of the patch antenna, which leads to a good skirt selectivity. To suppress undesired E-plane sidelobes caused by the out-of-phase currents of TM 03 mode, the H-plane slots are etched at the middle of the patch. Finally, the proposed TM 03 mode patch antenna with improved performance is designed and fabricated. The measured results achieve an impedance bandwidth of 3.74%, two radiation nulls near 4.65 and 5.8 GHz, a realized gain of 11.86 dBi at 5.1 GHz, as well as the sidelobe level of −11.54 dB, respectively.

In this paper, a novel power divider is designed with broadband phase reconfigurable capability. The phase difference of the power divider can be switched between 0° and 180°, and unequal power dividing ratio is achieved simultaneously. The broadband phase switchable characteristic is obtained by using a finite-width microstrip line swap and two switches, instead of relying on tightly coupled lines on high dielectric constant substrates. To validate the proposed circuit, a power divider with a 2:1 power dividing ratio at 1 GHz is designed and fabricated. Measured results show that the circuit achieves an operating bandwidth of 108%, while the return loss is better than 10 dB and the minimum in-band isolation is better than 22 dB.

Educational games, as the combined product of game and education, carry people's infinite desire for education and endless expectations for games. VR technology is considered be a “star” technology in education industry with practical applications after multimedia and internet. Moreover, speech recognition has received a lot of attention as one of the human-computer interaction methods nowadays. The speech recognition is based on VR game system designed in our research combines the two major trends of VR education and speech recognition. This VR game system is applied to the game systems and uses its speech recognition AI system to bring extra entertainment that based on education experience, besides with those traditional education modes; therefore, it has a bright development prospect.

With population aging at unprecedented levels, the number of people with cognitive disorders increases year by year. MMSE is an early detection method commonly used in medical institutions for patients with mild cognitive tendency. It is effective and reliable for the diagnosis and longitudinal evaluation of cognitive ability. However, many elders with cognitive predisposition are reluctant to go to the hospital for early examinations for privacy reasons. At the same time, early detection of cognitive disorders requires continuous observation. These impose burdens on some patients and medical units, and prompt us to develop and design electronic version of MMSE. We develop it with two purposes. First, it can be tested at any time, and upload test and analyze data simultaneously, further finding patients with cognitive predisposition in the early stage. Secondly, it helps hospitals to observe the population of cognitive predisposition, manage health data intelligently, and develop more effective treatment methods.

Dementia becomes more common as people live longer. The Mini-Mental State Examination (MMSE) is commonly used in medical to screen for dementia. MMSE has advantages that require no specialized equipment or training to use. It has both validity and reliability for diagnosis and longitudinal assessment of cognitive ability. However it still can't be conveniently used at home. Therefore, we have developed an electronic version of MMSE that can be used at home to assess cognitive ability and perform early detection.

In this paper, a residue number system (RNS)-based data embedding scheme is proposed, which converts secret information and image cover into residues. Since the residues represent the lowest levels of a continuous-tone image that human eyes are not sensitive to, the residues of cover can be considered as redundancy, which are replaced by the embedded residual data without introducing perceptible distortions. In addition, it provides additional security with a set of modulus. Simulation results demonstrate the proposed technique achieves higher imperceptibility and payload than other state-of-the-art techniques. Especially, when compared with the least significant bit (LSB) embedding technology, the cost of cover is reduced by 90.62% and the imperceptibility is increased by 4.87dB. Furthermore, in case of damage, the proposed method has a potential capability of lossless data recovery by the dynamic range adjustment technique.

A cross-mesh array antenna is presented for RF power harvesting in this paper. To enhance the reliability of harvester, producing multiple polarizations and multiple beams is an effective method in rectenna designs. Benefited from the ability of tilting beam angle of cross-mesh array antenna, four differently tilted beams with two linear polarizations can be simply generated by four isolated feed points without any beamforming network design. To demonstrate it, a prototype operating around 6 GHz is fabricated and measured.

This study presents a density based topology optimization method with prescribed input displacement boundary condition to design an elastomeric engine mount. The required stiffness values for the rubber mount have different requirements in two different loading directions. Various design analysis cases are performed to provide design guideline for investigating the effects of several design parameters on static stiffness of the elastomer mount. Simulation models are developed using the commercial finite element analysis program, LS-DYNA, to analyze the designs. One commercial elastomeric engine mount is selected as the validation example in this study. The analysis results show the simulated stiffness values agree well with experimental data. The design examples and results provided in this study can serve as the design reference for developing elastomer mounts with desired stiffness.

As for the industrial area, the control of VOCs emission has attracted a lot of attentions of researchers for several decades. Appropriate control strategies are fundamental tools to restrain the emission of VOCs. Based on empirical data extracted from the complete set of sources, control strategies for reducing VOCs emission process are proposed in this paper, which are obtained by performing numerical analyses. Empirical data are applied into the proof of theoretical results. Complete sources including industrial source, traffic source, living source and agricultural source are considered. The impacts of output value of general equipment in the industrial source, passenger throughput in traffic source, the size of population in living source and agricultural source on VOCs emissions are discussed. Additionally, control strategies for VOCs emissions are summarized by analyzing extremum of characteristic values obtained from sample sets. Finally, the features of various VOCs emission sources are concluded to guide the adjustment of treatment process, especially the treatment equipment of VOCs.

In this paper, a design of compact differentially-fed rectangular-ring microstrip antenna is proposed. In order to effectively excite the dominant TM 01 mode, two feed points are chosen with an offset from the symmetric axis. The simulation indicates that the resonance of the TM 01 mode occurs at 4.395 GHz, which results in a maximum gain of 7.53 dBi. Benefited from the differential signal operation, a relatively weak cross-polarization can be achieved.

We apply sequence-to-sequence model to mitigate the impact of speech recognition errors on open domain end-to-end dialog generation. We cast the task as a domain adaptation problem where ASR transcriptions and original text are in two different domains. In this paper, our proposed model includes two individual encoders for each domain data and make their hidden states similar to ensure the decoder predict the same dialog text. The method shows that the sequence-to-sequence model can learn the ASR transcriptions and original text pair having the same meaning and eliminate the speech recognition errors. Experimental results on Cornell movie dialog dataset demonstrate that the domain adaption system help the spoken dialog system generate more similar responses with the original text answers., Accepted at ASRU 2017

We propose a video analytic system to segment and track clouds for the purpose of solar irradiance prediction. Ground-based imaging sensors are used to monitor potential sun occlusions for solar irradiance drop prediction, which can be used to assist power ramp control strategies. Sky images are first rectified to remove fisheye artifacts. Cloud pixels are segmented and classified into low-to-high transparencies. Evolution of cloud boundaries are tracked using optical flow. Tracking extrapolation predicts future cloud movements and deformations as well as potential sun occlusions. To accurately estimate solar irradiance drop, we propose a novel scheme based on back-projecting the predicted sun occlusion onto the evolving cloud boundary to count for cloud transparency and irradiance drop estimation. Experimental results show that short-term solar irradiance drop is predictable with reasonable accuracy.

The propulsion system as the main power for the underwater robot will extremely influence its hydrodynamic performance. This paper focuses on a novel hybrid propulsion device for the third-generation spherical underwater robot (SUR III) with both vectored water-jet thrusters and propeller thrusters. To the limited space and mechanism, the hybrid propulsion device not only reserves the symmetric structure to maintain balance underwater, but also enhances the better property. And the diversity of the movement is also proposed for the different target as remote or hover. In order to analyze the hydrodynamic characteristics of the propeller, we establish the flow field with the multi-reference frame method to calculate the pressure and thrust of the propeller in ANSYS CFX. Finally, we set up the experimental system with a 6-DOFs load cell to measure the thrust of the propeller. Comparing the simulation and experiment results, the simulation thrust error is less than 8.5%. Meanwhile the power of the propeller thruster is 22.5% better than the water-jet thruster.

A novel rectenna based on the grid-array antenna is presented in this paper. The grid-array antenna provides the feasibility to tilt the radiation beam. Two isolated feed points are designed in the grid-array antenna so as to generate two different beams separately. In this way, the rectenna can harvest more wireless RF energy from different directions. Finally, a prototype is fabricated and measured. Under the power density of 1 μW/cm2, it can produce over 50 μW dc power when the incident angle of incoming wave ranges from −10° to 10° at the E-plane.

This study presents a portion of the results for the four-year Taiwan National Science Council Project, Development and Assessment of Science Courses Integrating Indigenous Culture and Collaborative Problem Solving(CPS). The research team has already conducted three years of teaching experiments at Atayal Rainbow Elementary School in Nan'ao Atayal tribes. Results demonstrate trained Atayal students displayed significantly better spatial concepts and abilities than untrained Atayal tribal students after taking the courses in this project.

This study consists of the partial result of the High Scope Plan "The Study of the Teaching of Dig-itized Project-Based Learning in the Atayal Indige-nous High School" of the Ministry of Science and Technology, ROC. This is a quantitative study. The participants were 24 indigenous high school students in Yilan. Through the teaching procedures, the students' visual design abilities of cultural and creative goods in 3D printing experience were studied. The students had best performance in "novelty" and "singularity".

Efficiently managing resources and improving throughput in a large-scale cluster has become a crucial problem with the explosion of data processing applications in recent years. Hadoop YARN and Mesos, as two universal resource management platforms, have been widely adopted in the commodity cluster for co-deploying multiple data processing frameworks, such as Hadoop MapReduce and Apache Spark. However, in the existing resource management, a certain amount of resources are exclusively allocated to a running task and can only be re-assigned after that task is completed. This exclusive mode unfortunately leads to a potential problem that may underutilize the cluster resources and degrade system performance. To address this issue, we propose a novel opportunistic and efficient resource allocation approach, named OpERA, which breaks the barriers among the encapsulated resource containers by leveraging the knowledge of actual runtime resource utilizations to re-assign opportunistic available resources to the pending tasks. We implement and evaluate OpERA in Hadoop YARN v2.5. Our experimental results show that OpERA significantly reduces the average job execution time and increases the resource (CPU and memory) utilizations.

In this paper, a compact rectenna with two closely spaced elements is proposed for the low-power energy harvesting application. To compensate the low RF-to-dc conversion efficiency of rectifier due to the low input power, an asymmetric configuration of antenna is designed, which achieves 7.3-dBi gain at 2.45 GHz. The rectifier is further embedded in the antenna without any additional circuit area. A prototype with a compact area of 11.5 mm × 60.0 mm is designed and evaluated with an incident power density of 1 μW/cm2. The measured results show that the proposed rectenna can produce 26.1-μW dc power with over 40% RF-to-dc conversion efficiency.