An Empirical Study on the Interaction Capability of Arm Stretching

Body-based motion gestures have been gaining popularity in designing interactive systems. However, theories and design guidelines on the use of body movement in design have not been fully evaluated. This article investigates human ability to perform discrete target selection tasks using stretching action through two controlled experiments. The experimental results indicate that: (1) the range of the discrete levels of depth that users can easily discriminate with arm stretching is up to 16 with full visual feedback, but is down to 4 without the feedback; (2) dwelling, a gesture with keeping a hand motionless and the cursor within a target area for a certain amount of time, may be the best gesture for confirmation command; (3) full visual feedback can improve the user performance; and (4) arm stretching action can be modeled using Fitts’ law. We also discuss the design potentials for Stretch Widgets based on these results.     

A weighted ADI scheme with variable time steps for diffusion-wave equations

Calcolo - We present a local projection stabilization for the virtual element discretization of the nonlinear convection–diffusion–reaction equation. We consider a...     

Regularization parameter estimation for feedforward neural networks

Under the framework of the Kullback-Leibler (KL) distance, we show that a particular case of Gaussian probability function for feedforward neural networks (NNs) reduces into the first-order Tikhonov regularizer. The smooth parameter in kernel density estimation plays the role of regularization parameter. Under some approximations, an estimation formula is derived for estimating regularization parameters based on training data sets. The similarity and difference of the obtained results are compared with other work. Experimental results show that the estimation formula works well in sparse and small training sample cases.     

Geostatistical downscaling of AMSR2 precipitation with COMS infrared observations

This article presents a geostatistical approach for downscaling precipitation products from passive microwave satellites with geostationary Meteorological Satellite observations. More precisely, the Advanced Microwave Scanning Radiometer 2 (AMSR2) precipitation (daily level 3 product) with 0.25° spatial resolution and the Communication, Ocean and Meteorological Satellite (COMS) infrared (IR) data with 5 km spatial resolution were used for the downscaling experiment over the Korean peninsula. Brightness temperature data observed at COMS IR 1 and water vapour channels were incorporated for downscaling via area-to-point residual Kriging with non-linear regression. The evaluation results with densely sampled Automatic Weather Station data revealed that incorporating the COMS IR observations with the AMSR2 precipitation showed similar error statistics to those of the AMSR2 precipitation because of the limitations of IR observations themselves and the inherent errors of the AMSR2 precipitation product over land. However, the area-based evaluation using information entropy indicated that incorporating the COMS observations resulted in more detailed spatial variation in the final product than direct downscaling of the AMSR2 precipitation. In addition, local precipitation patterns could be captured when there were regions with missing precipitation values in the AMSR2 product. Consequently, the downscaling result is useful for understanding the local precipitation patterns with an accuracy similar to that provided by microwave satellite observations. It is also suggested that the spatial variability in the downscaling result and errors in input low-resolution data should be considered when downscaling coarse resolution data using fine resolution auxiliary variables.     

Design of high performance linear feedback laws for operation that extends into the nonlinear region of AMB systems

Existing active magnetic bearings (AMBs) operate in the linear region of the magnetic material flux density, which limits the utilization of the bearing capacity. In order to increase the utilization of the bearing capacity and enhance the performance of the AMB system, this paper develops a method for designing high performance linear feedback laws. The resulting feedback laws allow the AMB to operate in its nonlinear region and hence improve the closed-loop performance. We first establish an approximate nonlinear AMB current force response model, and place this nonlinear curve inside a sector formed by two piecewise linear lines. Based on the linear line segments in these two piecewise linear lines, we determine the maximum disturbance that can be tolerated by solving an optimization problem with linear matrix inequality (LMI) constraints. For a given level of disturbance under the maximum tolerable disturbance, we formulate and solve the problem of designing the linear feedback that achieves the highest level of disturbance rejection as another LMI problem. Both $L_2$ disturbances and $L_\infty$ disturbances are considered. Finally, we illustrate our design by both simulation and experimental results.     

目标提取与哈希机制的多标签图像检索

目的 哈希是大规模图像检索的有效方法。为提高检索精度,哈希码应保留语义信息。图像之间越相似,其哈希码也应越接近。现有方法首先提取描述图像整体的特征,然后生成哈希码。这种方法不能精确地描述图像包含的多个目标,限制了多标签图像检索的精度。为此提出一种基于卷积神经网络和目标提取的哈希生成方法。方法 首先提取图像中可能包含目标的一系列区域,然后用深度卷积神经网络提取每个区域的特征并进行融合,通过生成一组特征来刻画图像中的每个目标,最后再产生整幅图像的哈希码。采用Triplet Loss的训练方法,使得哈希码尽可能保留语义信息。结果 在VOC2012、Flickr25K和NUSWIDE数据集上进行多标签图像检索。在NDCG(normalized discounted cumulative gain)性能指标上,当返回图像数量为 1 000时,对于VOC2012,本文方法相对于DSRH(deep semantic ranking hashing)方法提高24个百分点,相对于ITQ-CCA(iterative quantization-canonical correlation analysis)方法能提高36个百分点;对于Flickr25,本文方法比DSRH方法能提高2个左右的百分点;对于NUSWIDE,本文方法相对于DSRH方法能提高4个左右的百分点。对于平均检索准确度,本文方法在NUSWIDE和Flickr25上能提高25个百分点。根据多项评价指标可以看出,本文方法能以更细粒度来精确地描述图像,显著提高了多标签图像检索的性能。结论 本文新的特征学习模型,对图像进行细粒度特征编码是一种可行的方法,能够有效提高数据集的检索性能。     

Exploring cell tower data dumps for supervised learning-based point-of-interest prediction (industrial paper)

Exploring massive mobile data for location-based services becomes one of the key challenges in mobile data mining. In this paper, we investigate a problem of finding a correlation between the collective behavior of mobile users and the distribution of points of interest (POIs) in a city. Specifically, we use large-scale cell tower data dumps collected from cell towers and POIs extracted from a popular social network service, Weibo. Our objective is to make use of the data from these two different types of sources to build a model for predicting the POI densities of different regions in the covered area. An application domain that may benefit from our research is a business recommendation application, where a prediction result can be used as a recommendation for opening a new store/branch. The crux of our contribution is the method of representing the collective behavior of mobile users as a histogram of connection counts over a period of time in each region. This representation ultimately enables us to apply a supervised learning algorithm to our problem in order to train a POI prediction model using the POI data set as the ground truth. We studied 12 state-of-the-art classification and regression algorithms; experimental results demonstrate the feasibility and effectiveness of the proposed method.     

基于改进卷积注意力模块与残差结构的SSD网络

SSD(Single Shot Multibox Detector)是一种基于卷积神经网络的单阶检测算法,相比双阶检测算法,它在保证一定精度的同时显著提高了检测速度,但仍难以满足很多实际应用,尤其是在小目标检测任务中,检测精度更是难以满足需求.针对该不足,文中提出了一种基于改进残差结构与卷积注意力模块的特征提取网络Re...