Search

Temporal action localization (TAL) is a task of identifying a set of actions in a video, which involves localizing the start and end frames and classifying each action instance. Existing methods have addressed this task by using predefined anchor windows or heuristic bottom-up boundary-matching strategies, which are major bottlenecks in inference time. Additionally, the main challenge is the inability to capture long-range actions due to a lack of global contextual information. In this paper, we present a novel anchor-free framework, referred to as HTNet, which predicts a set of triplets from a video based on a Transformer architecture. After the prediction of coarse boundaries, we refine it through a background feature sampling (BFS) module and hierarchical Transformers, which enables our model to aggregate global contextual information and effectively exploit the inherent semantic relationships in a video. We demonstrate how our method localizes accurate action instances and achieves state-of-the-art performance on two TAL benchmark datasets: THUMOS14 and ActivityNet 1.3., Comment: 6 pages

Neural prostheses are systems that interact with the human nervous system to recover function lost as a result of disease by recording or stimulating neural signals. Neural implants have a variety of components, but among them, designing highly efficient wireless power transmission systems is particularly important. However, there is a problem that it is difficult to design a wireless power transmission system with high transmission efficiency because power transmission efficiency is affected by various coil parameters. In addition, nerve implants are mainly designed based on polymer substrates, but studies have not been conducted on the process of manufacturing coils based on polymer substrates. In this study, to overcome the mentioned problems, an easy and fast manufacturing process was proposed before designing a precise coil on a polymer substrate. Coil was made by manually winding the 200um copper wire, and the coil was packaged using COC. The coil feasibility test was conducted after fabrication to evaluate the performance of fabrication process. It did not fit the expected frequency because the wire was manually wound to make a small size coil. However, the wire wound coil is expected to be more efficient than the coil made through the PCB process, and if the sophistication is supplemented, it can be used as a highly efficient coil for wireless power transmission.

Object localization in video is to predict the location and image boundaries of objects of interest in sequential scenes. Despite numerous methods being developed for the task, there are still challenging issues, such as labor-intensive data preparation. In this paper, we propose a patch-wise approach with weak supervision to resolve those issues in the object localization. We first train an patch-wise object classifier based on convolutional neural network with simple labeling about object classes, instead of the bounding box annotation. Then, the object regions are estimated using the class activation maps of the classifier for each patch. The patch-wise classifier can learn more relevant features of objects from the patches containing various parts of them. In addition, background patches for weakly-supervised learning can be easily prepared. Experiments using the visual object tracking challenge data set showed that the patch-wise weakly supervised approach is effective in the object localization in video.

We demonstrated a highly reliable buried-gate saddle-fin cell-transistor (cell-TR) using silicon migration technique of hydrogen (H 2 ) annealing after a dry etch to form the saddle-fin in a fully integrated 2y-nm 4Gb DRAM. It clearly shows a reduction in interface trap density with highly enhanced variable-retention-time (VRT) and Row-Hammering immunity.

We consider a network that the transmitters are located as Poisson point process and each has its receiver at a given distance. Additionally, a packet arrives at the transmitter following Bernoulli distribution and departs with medium access probability. The success of transmission is only when the SINR is greater than a given threshold. In this situation, we improve the previous work by considering varying multi-traffic class and general kind of fading in the network. With these additional assumptions for practical situation, we investigate the stability condition, success probability, and the mean delay of a packet.

In this paper, we tackle the physical-layer security issue in MISO wiretap channel considering the pilot-assisted channel estimation at the receiver side. Unlike the conventional schemes, we introduce the cooperative jamming scheme that transmits artificial noise (AN) only on downlink pilot transmission phase. In order to disrupt the channel estimation at the eavesdropper (Eve) effectively, an optimal beamforming at the cooperative jammer (CJ) was obtained by means of difference-of-convex (DC) algorithm and rank relaxation. In addition, with focus on the energy budget limitation issue at the CJ, energy efficiency at the CJ is verified based on the circuit power consumption model. Numerical results provide a performance comparison with a conventional scheme, which transmits AN only in data transmission phase.

Apache Spark framework, which is the implementation of Resilient Distributed Datasets(RDD), is used instead of MapReduce on recent data processing models of Hadoop ecosystem. In this paper, we evaluated the performance and resource usage of real world workloads on scale-up and scale-out clusters using the in-memory caching feature of Spark framework. In our experiments, scale-up processed data more efficiently than scaleout in write intensive workloads such as Sort and Scan, whereas scale-out had strength in those utilizing iterative algorithms such as Join, Pagerank and KMeans. Considering the efficiency in physical factors including performance per watt and the physical space each occupies, we show that it is more advantages to use scale up cluster than scale out.

In this paper, we investigate techniques to effectively orchestrate HDFS in-memory caching for Hadoop. We first evaluate a degree of benefit which each of various MapReduce applications can get from in-memory caching, i.e. cache affinity. We then propose an adaptive cache local scheduling algorithm that adaptively adjusts the waiting time of a MapReduce job in a queue for a cache local node. We set the waiting time to be proportional to the percentage of cached input data for the job. We also develop a cache affinity cache replacement algorithm that determines which block is cached and evicted based on the cache affinity of applications. Using various workloads consisting of multiple MapReduce applications, we conduct experimental study to demonstrate the effects of the proposed in-memory orchestration techniques. Our experimental results show that our enhanced Hadoop in-memory caching scheme improves the performance of the MapReduce workloads up to 18% and 10% against Hadoop that disables and enables HDFS in-memory caching, respectively.

Loosely coupled applications composed of a potentially very large number (from tens of thousands to even billions) of tasks are commonly used in High-Throughput Computing (HTC) and Many-Task Computing (MTC) paradigms. To efficiently execute large-scale computations which can exceed the capability in a single type of computing resources within expected time, we should be able to effectively integrate resources from Heterogeneous Distributed Computing (HDC) systems such as Clusters, Grids, and Clouds. In this paper, we quantitatively analyze the performance of three different real scientific applications consisting of many tasks on top of HDC systems based on a Partnership of Distributed Computing Clusters, Grids, and Clouds to show practical issues that normal scientific users can face during the course of leveraging these systems. Our experimental results show that the performance of a loosely coupled application can be significantly affected by the characteristics of a HDC system, along with hardware specification of a node, and their impacts on the performance can vary widely depending on the resource usage pattern of each application. Throughout our extensive performance study with representative HDC systems and real scientific applications, we aim to give an insight to the research community on design and implementation of a next generation middleware system that can intelligently support large-scale loosely coupled applications by considering both of resource and application perspectives.

This paper proposes an approach to develop reduced-order electrochemical battery models for controls. The reduced-order models are derived from detailed electrochemical battery models, which combine ionic and electronic processes in the level of active material particles, through a significant model reduction with the order of hundreds or more. The reduction is possible through the understanding of electrochemical process in battery cells. The developed models can capture most of key electrochemical dynamics in a cell with a few state variables. The model is validated using real-world vehicle test data, which consist of both charge depleting (CD) and charge sustaining (CS) driving modes. The models were implemented and tested in a battery management system (BMS).

Model-based thermal management enables sophisticated battery temperature control due to the availability of temperature information among and inside cells. This paper proposes a reduced-order battery thermal modeling approach by combining the pre-calculated temperature distribution in steady state and the approximated transient responses from multi-dimensional full-order thermal models. The number of states is reduced by an order of hundreds, and the developed model can predict battery thermal dynamics with errors within 0.2 °C compared to the full order model simulation results under a real-world driving cycle.

Recent emerging applications from a wide range of scientific domains often require a very large number of loosely coupled tasks to be efficiently processed. To support such applications effectively, all the available resources from different types of computing platforms such as supercomputers, grids, and clouds need to be utilized. However, exploiting heterogeneous resources from the platforms for multiple loosely coupled many-task applications is challenging, since the performance of an application can vary significantly depending on which platform is used to run it, and which applications co-run in the same node with it. In this paper, we analyze the platform and co-runner affinities of many-task applications in distributed computing platforms. We perform a comprehensive experimental study using four different platforms, and five many-task applications. We then present a two-level scheduling algorithm, which distributes the resources of different platforms to each application based on the platform affinity in the first level, and maps tasks of the applications to computing nodes based on the co-runner affinity for each platform in the second level. Finally, we evaluate the performance of our scheduling algorithm, using a trace-based simulator. Our simulation results demonstrate that our scheduling algorithm can improve the performance up to 30.0%, compared to a baseline scheduling algorithm.

This paper proposes an equivalent circuit model of 3-D DRAM cell transistors with recess gate and saddle fin structure for the first time. The model effectively characterize the sub-threshold and off margin behavior of the scaled DRAM cell transistor by considering the parasitic sub-channel and vertical transistor components into account. TCAD simulation and experimental data have confirmed the accuracy of the model. With the analysis made, we suggest a set of improvement method for the off margin characteristics engineering. These methods are believed to lead the continuous DRAM scaling, down to sub-10nm technology node.

Wireless sensor network (WSN) localization has drawn attention in a broad spectrum of applications for security, disaster response, monitoring, surveillance and tactical system. The deployment of a large number of low-cost, low-power and multi-functional sensors is preferred, which provides an exceptional localization performance via cooperative communication. In this paper, we consider a cooperative localization in WSNs consisting of multiple anchor nodes (ANs) equipped with a linear antenna array of M elements, a single-antenna target and a central unit (CU). We focus on a synchronous network and all the channels are supposed to be frequency-flat. Our objective is to evaluate the performance of localization technique for the target's position that jointly exploits both Time-of-Arrival (TOA) and Angle-of-Arrival (AOA) measurements. We derive the Cramer-Rao bound (CRB) for a joint TOA/AOA-based localization. Additionally, for the performance comparison, the CRBs of solely TOA- and AOA-based localizations are also derived. By numerical results, our analysis is verified and it is observed that the joint TOA/AOA-based localization has a superior performance than either TOA- or AOA-based localization.

In the cooperative positioning system (CPS), members of a group communicate each other and share their information to localize their position. The advantage of this system is that all the elements in the CPS can obtain their position collaboratively even if only a few members of the group get their position from GPS. This paper presents a low-cost CPS using two-way ranging (TWR) measurements and GPS. Using chirp spread spectrum ultra wideband (CSS-UWB) modulation scheme, precise TWR measurements with less than 1m error can be obtained and they are used to solve the position of members that cannot acquire GPS signal. This paper also analyzes the error propagation in the CPS. From the ranging covariance matrix, the location covariance matrix is derived to analyze the position error propagation. Simulations with various scenarios show that the geometry of members is the most important factor in the error propagation.

This paper describes the design of a wrist rotation rehabilitation robot based on force measurements for severe stroke patients lying in bed wards or at home which allows them to receive wrist rehabilitation exercise. The manufactured six-axis force/torque sensor which can detect three directional force Fx, Fy, Fz and three directional torque Tx, Ty, Tz, was attached to the developed rehabilitation robot and allows the rehabilitation robot to measure a rotational force (Tz) exerted on a wrist, the signal force Fx, Fy, Fz and torque Tx, Ty which in turn allows the device to be used safely. The robot was designed and manufactured for severe stroke patients lying in bed, and a robot program was manufactured to perform a wrist rotation flexibility rehabilitation exercise. The results of a characteristics test for the developed rehabilitation robot showed that it was safely operated while the wrist rotation flexibility rehabilitation exercise was performed. Therefore, it is thought that the developed rehabilitation robot can be safely used with severe stroke patients.

A magnetic contactor (MC) using permanent magnets has a critical problem. When an unintended external force, such as an earthquake or a shock, is applied to the mover, the MC can be closed unexpectedly. To address this problem, the inter-locking system is proposed for the permanent magnet (PM) type MC in this research. Hence, the reliability of the PM type MC is improved significantly via the proposed inter-locking system.

We propose a meanshift segmentation based inverse colorization method for image compression. The encoder makes use of the meanshift segmentation algorithm in automatically selecting the representative pixels from the original image from which the colored image is reconstructed by the decoder. Using the modes of the clustered regions as the representative pixels, the compression rate becomes high and the reconstructed image has good visual quality.

The carrier-smoothed GPS (CSGPS) is a method of estimating a precise trajectory with carrier phase measurements. The integer ambiguities in carrier phase measurements are estimated with code measurements. The CSGPS can be used to provide precise trajectory for high-precision applications. In case of the cooperative simultaneous localization and mapping (CSLAM), localization is effectively accomplished by using CSGPS. However, visibility change of GPS satellite vehicle (SV) in tracking or cycle slips make discontinuity in the trajectory. Therefore, dead reckoning (DR) sensor should be utilized to compensate discontinuities. This paper analyses the performance of CSGPS for an hour without bias compensation. The structure of a loosely coupled Kalman filter that combines CSGPS and DR measurements is also shown.

Ultra Wideband (UWB) techniques are utilized to localize mobile robots in indoor environment. Although there are many methods, such as IR-UWB, CSS-UWB, and AWGN-UWB, it is hard to make it smaller. CSS-UWB technique is a viable solution, since it has a chip solution. By utilizing chirp signal, ranging counter, and SDS-TWR technique, CSS-UWB acquires under-meter ranging measurements. To solve position with these measurements in severe indoor condition, robust location algorithm should be exploited. Gauss-Newton (GN) method shows good performance compared to other algorithms. However, GN method has a chance to diverge in poor geometry condition. Moreover, different anchor heights could cause divergence problem in three-dimensional condition. The quadratic corrected least square (QCLS) method is introduced to alleviate this divergence problem. Not only classical QCLS method, but also other methods such as modified QCLS (MQCLS) are introduced. The performances of these algorithms are analyzed by both simulation and experimental results. MQCLS algorithm shows better performance for the edge of the geometry.

We propose transceiver algorithms to manage interference in cognitive radio networks where cognitive users are equipped with multiple antennas, which is referred to as spatial sensing technique. This work considers designs of precoding and decoding matrices for spatial sensing to protect a primary network from interference of the secondary network and to remove interference by primary transmissions at the secondary receiver. As an extensive work of [1], which studied a case of a zero-forcing (ZF) receiver, in this paper, we investigate a transceiver design with a minimum mean square error (MMSE) receiver. In the design with a ZF receiver, we assumed that the interference by primary transmission is perfectly removed at the secondary receiver. In this work, to further improve the performance of secondary networks, we propose using an MMSE receiver. We show that an iterative algorithm is necessary. Numerical results confirm that the design with an MMSE receiver outperforms the design with a ZF receiver.

We have designed and fabricated surface plasmon resonance (SPR) sensor with metallic nano-structure to improve sensitivity. We have fabricated the metallic nano particles with rapid thermal annealing process to realize the proposed nano-plasmonics SPR sensor. In experiments, we confirmed the sensitivity of the proposed nano-plasmonics SPR sensor is 93.75 degree/RIU while that of conventional SPR sensor is 84.76 degree/RIU.

This paper proposes an approach to characterize the plug-in hybrid electric vehicle (PHEV) behavior for assessing the impact of PHEVs on the grid based on naturalistic driving information. PHEV behavior is captured by response surfaces of the amount for electric energy consumption and the amount of fuel consumption at arrival with respect to the initial SOC and the driving distance. Response surfaces are created from detailed PHEV simulation results using representative cycles synthesized based on naturalistic driving data. The proposed approach enables computationally efficient prediction of the PHEV impact on the grid with large data-sets.

Knowledge of the internal temperatures of an electric machine under real-time operating conditions would be extremely useful in order to determine its torque capabilities. This knowledge is also useful for full-scale electric vehicle simulation and optimization. In this paper we present a technique for developing computationally-efficient thermal models for electric machines that can be used for real-time thermal observers and vehicle-level simulation and optimization. The technique is based upon simulating the eigenmode of the system as determined by finite element analysis. The order of the model is then dramatically reduced in two ways. First, the extent of excitation of each mode is calculated, and only eigenmodes that are significantly excited are included in the dynamic model; other eigenmodes are treated as static modes. Second, only a few “hot spots” in various regions are chosen. The result is a thermal model that can accurately model internal temperatures of the machine while requiring the modeling of only a handful of states. Such a model can be used in vehicle simulations, or for real-time observers in actual vehicles. The computation time of the model presented in this paper is dramatically reduced compared with a typical full-order finite element model.

Accurate prediction of the battery electrochemical dynamics is important to avoid undesired battery operation under aggressive driving. This paper proposes a battery power management strategy considering Li-ion concentration in the electrodes to prevent excessive battery charging and discharging. The proposed approach adjusts the allowable battery power limits through the feedback of the estimated electrode-averaged Li-ion concentration information. An advanced hybrid electric vehicle (HEV) power split strategy is constructed implementing a Li-ion battery model with electrochemical diffusion dynamics to capture the battery dynamic behavior under transients. A novel contribution arises from the implementation of an extended Kalman filter (EKF) using uneven discretization of the particle radius for fast and accurate prediction of the Lithium intercalation dynamics. The control design modifies the allowable battery power limit used in the supervisory controller, thus, maintaining low complexity of the control structure.

Driving cycles play an important role in analysis and design of hybrid propulsion systems. Their role is magnified in case of Plug-In Hybrids, since driving patterns have a strong impact on specific energy consumption, ability of the vehicle to provide pure electric operation, range of charge depleting operation, and finally assessments of the state-of-charge (SOC) at the beginning of charging. Certification schedules are commonly used, but in case of PHEV standard cycles are repeated many times to study the charge depleting operation. While this is useful for comparisons with results in literature, it is not representative of real-world driving. Therefore, characterizing the naturalistic driving patterns and generating representative real-world schedules is essential for PHEV design and control work, studies of technology adoption by real consumers, and assessments of the impact on the grid. This work presents analysis of naturalistic driving data collected in a Midwest region of US, followed by a procedure for synthesizing real-world driving cycles based on that data. The characteristic of the real-world cycles are captured with transition probability matrices (TPMs), and a regression analysis is applied to select significant explanatory variables for determining the most representative synthetic cycles. Validation indicates the ability to generate representative real-world driving cycles for any arbitrary driving distance by combining a stochastic process and statistical methodology.

There have been tremendous demands on portable electronic devices since last decade and board level drop performance has become one of the major concerns in microelectronic ball grid array (BGA) package. Most studies focused on material selection of solder joints to improve drop performance [1]. However, the risk for early failure in drop impact reliability still exists because it is sensitive to several factors such as substrate supplier, reflow conditions, loading condition and aging[2,3]. In this paper, mild aging effect on drop performance was investigated to reduce early failure risk for lead free solder joints. From this study, for the case of relatively poor performance and interface dominant failure mode, mild aging could improve initial drop life with brittle-to-ductile transition.

In this paper, we propose a fast motion estimation algorithm to reduce computational load of the FS (full search) algorithm. We reduce only unnecessary computations which doesn't affect predicted images from the motion vector. To do that, we use normalized dithering matching scan to get uniform distribution of partial distortion by which we can reduce only unnecessary computation significantly. Additionally, we remove unlike candidate vectors faster by using multiply of LOG value to threshold of matching error. Additionally, we further reduce unnecessary computation for motion estimation according to distribution of motion vectors. Our algorithm reduces about 60% of computations for block matching error compared with the conventional PDE (partial distortion elimination) algorithm almost without degradation of prediction quality.

Due to cell planning in a cellular communication network, it is difficult for a mobile user in a cell to get ranging measurements from other neighboring cells. Consequently, it is necessary to enhance signal sensitivity using long integration techniques so that the user position is obtained by trilateration. In order to maximize integration gain in coherent/noncoherent hybrid long integration, it is desirable to make SNR (signal to noise ratio) after coherent integration become as large as possible. Because SIR (signal to interference ratio) is not improved by the long integration, however, it is needed to mitigate CCI (Co-Channel Interference) so that the resultant hearability can be enhanced for weak signals. This paper presents a hybrid long integration scheme in presence of CCI. SNIR (signal to noise plus interference ratio) is employed to explain the loss induced by interference as well as noise. Together with interference cancellation and residual frequency estimation technique, design procedure of hybrid long integration is proposed to maximize SNIR. Simulation results show that hearability is greatly improved at most of cell coverage when the proposed method is used.

In mobile WiMAX, geo-location is possible at the PSS (personal subscriber station) by using the preamble signals broadcasted by all the RAS (Radio Access Station) simultaneously. Using relative arrival times of signals, user location can be obtained by TDoA (Time Difference of Arrival) positioning. However, sufficient number of RAS should be detected to obtain the location in this method. Since mobile WiMAX system employs cellular network structure, the number of RAS that PSS can detect is limited due to low SNR and CCI (co-channel interference) by neighboring RAS. To detect sufficient number of RAS at the PSS, hearability should be enhanced by increasing SNR (signal to noise ratio) and SIR (signal to interference ratio), which can be achieved using long integration and interference cancellation. In multipath environment, however, interference cancellation does not remove the CCI completely. Therefore, multipath mitigation and interference cancellation method should be merged and solved together. In this paper, detection performance after interference cancellation is analyzed using mobile WiMAX geo-location system. Simulation results show that trilateration is possible in most area of the cell coverage after interference cancellation in multipath environment, except at the vicinity of the RAS.

The Genetic Algorithm(GA) is suggested to be combined with the Adaptive Frequency Sampling(AFS) technique to find the decoupling capacitors for suppressing the resonances of the large power/ground plane(PGP)s. In more detail, the AFS is used to characterize each of the divided regions of the PGPs and compound of the AFS results are used to evaluate the fitness or cost function in each GA iterative step.

This study aims to finish effectively the thin magnesium (AZ31) alloy plate that is usually very difficult to be polished by the conventional cutting processes because of their explosive and fragility properties. Reaction components on the workpiece surface roughness were measured after the magnetic abrasive polishing of the magnesium alloy plate. Taguchi experimental design method was adopted for evaluating the effect of the process parameters on the improvement of the surface roughness by the magnetic abrasive polishing. The process parameters were the applied current to the inductor, the working gap between the workpiece and the inductor, the rotational speed and the feed speed. Optimal conditions for better surface roughness were determined using the S/N ratio. As a result, it was seen that the magnetic abrasive polishing was very useful for finishing the thin magnesium alloy plate.