A Method for Motion Abstraction Based on Haptic Information Directionality and an Application to Haptic Motion Display System

Robots working in the human environment have been researched in the field of motion control. For the next-generation robot, human and robot interaction technologies are needed. In particular, learning and displaying of human haptic motion are important. Therefore, the authors have proposed a method for abstracting haptic motion and designed a haptic motion display system. The motion abstraction method divides a measured motion to each action from the point of force and position. Then, action modes have been defined for expressing each divided action. Action modes are expressing force directionality or position directionality. By utilizing the proposed motion abstraction method, various kinds of human motion are abstracted as action modes. The designed haptic motion display system is trying to show these various kinds of human motion. This paper defines human action modes and environmental action modes from action modes. Human action modes are expressing human action force directionality, while environmental action modes are expressing environmental position response directionality. Furthermore, the haptic motion display system is redesigned. This system is redesigned based on human action modes and environmental action modes. The validity of the proposed method is confirmed by the experimental results.     

A Broad Learning-Driven Network Traffic Analysis System Based on Fog Computing Paradigm

The development of communication technologies which support traffic-intensive applications presents new challenges in designing a real-time traffic analysis architecture and an accurate method that suitable for a wide variety of traffic types.Current traffic analysis methods are executed on the cloud,which needs to upload the traffic data.Fog computing is a more promising way to save bandwidth resources by offloading these tasks to the fog nodes.However,traffic analysis models based on traditional machine learning need to retrain all traffic data when updating the trained model,which are not suitable for fog computing due to the poor computing power.In this study,we design a novel fog computing based traffic analysis system using broad learning.For one thing,fog computing can provide a distributed architecture for saving the bandwidth resources.For another,we use the broad learning to incrementally train the traffic data,which is more suitable for fog computing because it can support incremental updates of models without retraining all data.We implement our system on the Raspberry Pi,and experimental results show that we have a 98%probability to accurately identify these traffic data.Moreover,our method has a faster training speed compared with Convolutional Neural Network(CNN).     

基于主动学习和SVM方法的网络协议识别技术

针对未知网络协议数据流的获取与标记工作主要依赖于领域专家。然而,样本数据量的增加会导致人工成本超过实际负荷。提出了一种新颖的未知网络协议识别方法。该方法基于主动学习算法,仅依靠原始网络数据流的载荷部分实现对未知网络协议的有效识别。实验结果表明,采用该方法设计的识别系统在保证识别准确率和召回率的前提下,能够有效地降低学习过程中标记的样本数目,更适用于实际的网络应用环境。     

基于受试者工作特征pAUC优化的人脸识别系统

基于深度学习的人脸识别技术在大量应用场景中表现出优于传统方法的性能,它们的损失函数大致可分为2类：基于验证的和基于辨识的。验证型损失函数符合开集人脸识别的流程,但实施过程比较困难。因此目前性能较优的人脸识别算法都是基于辨识型损失而设计的,通常由softmax输出单元和交叉熵损失构成,但辨识型损失并没有将训练过程与评估过程统一起来。本文针对开集人脸识别任务提出一种新的验证型损失函数,即最大化受试者工作特征(ROC)曲线下的部分面积(pAUC);同时还提出一种类中心学习策略提高训练效率,使提出的验证型损失和辨识型损失有较强的可比性。在5个大规模非限定环境下的人脸数据集上的实验结果表明,提出的方法和目前性能最优的人脸识别方法相比,具有很强的竞争性。     

Friction modeling and compensation for haptic display based on support vector machine

As the presence of friction in a haptic display device seriously affects its performance, proper compensation of the frictional effects in such a device is of practical importance for advanced virtual reality applications where haptic display plays a critical role. This paper addresses the issue of friction modeling and compensation for haptic control system designs. A new method based on the Support Vector Machine (SVM) is developed in a controller design based on a two-port network to achieve accurate haptic display. The approximation model of friction is established offline through SVM learning and is used for online feed forward friction compensation. The advantages of this novel method are demonstrated through the experiments performed.     

基于贝叶斯深度学习的一维雷达有源干扰信号识别方法

雷达干扰信号准确识别是雷达抗干扰的前提,对于雷达生存至关重要。针对传统雷达干扰信号识别方法需要繁琐的分析计算提取特征,通用性差,泛化能力弱,难以适应复杂的雷达工作环境问题。本文考虑无需人工提取特征信息且具有较好的分类识别效果的深度学习网络。考虑到传统的深度学习网络由于使用点估计方式,不能够很好的衡量预测结果中的不确定性,本文提出了一种基于贝叶斯深度学习的干扰识别方法。首先,通过概率建模代替网络参数模型的点估计,解决了不确定性随机数据引起的网络过拟合问题。其次,考虑有效利用雷达回波信号的时序特性设计了LSTM层,同时解决训练过程中的梯度消失问题。基于线性调频雷达有源干扰实测数据完成了网络训练与测试,实验结果表明,引入贝叶斯方法可以在加快网络收敛速度的同时有效提高识别准确率。     

A nonsubsampled countourlet transform based CNN for real image denoising

The state of the art deep learning based denoising methods can achieve great denoising results. However, due to the lack of clean training data, the ground truth and noise level are unknown, traditional denoising methods are difficult to remove blind noise in general images. Furthermore, deep learning methods require specific noise levels to train the model, and specific models are built only deal with one noise level. In this paper, we propose a Nonsubsampled Countourlet Transform based convolutional network model (CTCNN) to deal with Gaussian noise and the noise of real images. The model is modified by U-Net, nonsubsampled Countourlet Transform (NSCT) and inverse NSCT are used to instead of sum pooling layer and up-convolution operation. NSCT can decrease the size of feature maps and preserve details of images without information loss. Different training strategies are adopted to train models in order to handle blinding noise such as underwater images which contain noise naturally. Simulation results show the proposed method is effective in standard images dataset and blind noisy images. The model we proposed has been compared with other state of the art denoising methods, and better subjective representation and PSNR improvement are obtained.     

A weighted-principal component regression method for the identification of physiologic systems

We introduce a system identification method based on weighted-principal component regression (WPCR). This approach aims to identify the dynamics in a linear time-invariant (LTI) model which may represent a resting physiologic system. It tackles the time-domain system identification problem by considering, asymptotically, frequency information inherent in the given data. By including in the model only dominant frequency components of the input signal(s), this method enables construction of candidate models that are specific to the data and facilitates a reduction in parameter estimation error when the signals are colored (as are most physiologic signals). Additionally, this method allows incorporation of preknowledge about the system through a weighting scheme. We present the method in the context of single-input and multi-input single-output systems operating in open-loop and closed-loop. In each scenario, we compare the WPCR method with conventional approaches and approaches that also build data-specific candidate models. Through both simulated and experimental data, we show that the WPCR method enables more accurate identification of the system impulse response function than the other methods when the input signal(s) is colored.     

深度学习下的散射成像:物理与数据联合建模优化（特邀）

为了利用被散射的光信号实现成像，越来越多的散射成像方法被提出。其中深度学习以其强大的数据表征和信息提取能力在散射成像领域发挥着重要的作用。相较于传统散射成像方法，基于深度学习的散射成像方法在成像速度、质量、信息维度等方面都有着巨大的优势。但是，模型训练、模型泛化等问题也制约着该方法的发展。因此，越来越多的研究将物理过程与基于数据驱动的方法进行联合建模，利用物理先验指导神经网络优化。相较于单纯的数据驱动方法而言，物理-数据联合建模的方法对数据量、神经网络参数量的依赖程度大大降低，在保证成像质量的前提下有效降低数据获取难度及对实验环境的要求。联合建模优化的方式实现了介质、目标类型等散射成像中关键节点的泛化。同时在训练过程方面，实现了从有监督到半监督再到无监督的训练优化过程迭代，不同模型和监督方式的提出大大提升了基于深度学习方法的训练效率，在降低对硬件和时间成本的同时，提升了基于深度学习的散射成像方法在非实验室场景应用的可能性。     

基于主动学习方法的网络流分类研究

提出了一种基于主动学习方法的网络流分类方法,采用主动学习技术提取少量高质量的训练样本进行建模.并提出了一种基于轮盘赌选择的样本筛选方法,能够有效避免已有主动学习方法中的早熟收敛现象.实验结果表明,其相对于已有的流识别方法,能够在仅依赖少量高质量训练样本的前提下,保证较高的识别正确率,更适用于现实网络环境.     

基于迁移学习的软件缺陷预测

传统软件缺陷预测方法在解决跨项目缺陷预测过程中适应能力不足,主要是因为源项目和目标项目之间存在不同的特征分布.为了解决这个问题,提出一种新的加权贝叶斯迁移学习算法,算法首先收集训练数据和测试数据的特征信息,然后计算特征差异,将不同项目数据之间差异转化为训练数据权重,最后基于这些权重数据建立预测模型.在8个开源项目数据集上进行实验比较,实验结果表明与其他方法相比本文方法显著提高跨项目缺陷预测性能.     

An optimal neural network process model for plasma etching

Neural network models of semiconductor processes have recently been shown to offer advantages in both accuracy and predictive ability over traditional statistical methods. However, model development is complicated by the fact that back-propagation neural networks contain several adjustable parameters whose optimal values are initially unknown. These include learning rate, initial weight range, momentum, and training tolerance, as well as the network architecture. The effect of these factors on network performance is investigated here by means of a D-optimal experiment. The goal is to determine how the factors impact network performance and to derive a set of parameters which optimize performance based on several criteria. The network responses optimized are learning capability, predictive capability, and training time. Learning and prediction accuracy are quantified by the experimental error of the model. The process modeled is polysilicon etching in a CCl ₄-based plasma. Statistical analysis of the experimental results reveals that learning capability and convergence speed depend mostly on the learning parameters, whereas prediction is controlled primarily by the number of hidden layer neurons. An optimal network structure and parameter set has been determined which minimizes learning error, prediction error, and training time individually as well as collectively     

Virtual environments for medical training: graphical and hapticsimulation of laparoscopic common bile duct exploration

We develop a computer-based training system to simulate laparoscopic procedures in virtual environments for medical training. The major hardware components of our system include a computer monitor to display visual interactions between 3D virtual models of organs and instruments together with a pair of force feedback devices interfaced with laparoscopic instruments to simulate haptic interactions. We simulate a surgical procedure that involves inserting a catheter into the cystic duct using a pair of laparoscopic forceps. This procedure is performed during laparoscopic cholecystectomy to search for gallstones in the common bile duct. Using the proposed system, the user can be trained to grasp and insert a flexible and freely moving catheter into the deformable cystic duct in virtual environments. The associated deformations are displayed on the computer screen and the reaction forces are fed back to the user through the force feedback devices. A hybrid modeling approach was developed to simulate the real-time visual and haptic interactions that take place between the forceps and the catheter, as well as the duct; and between the catheter and the duct     

Phantom-based multimodal interactions for medical education and training: the Munich knee joint simulator

Simulation environments based on virtual reality technologies can support medical education and training. In this paper, the novel approach of an "interactive phantom" is presented that allows a realistic display of haptic contact information typically generated when touching and moving human organs or segments. The key idea of the haptic interface is to attach passive phantom objects to a mechanical actuator. The phantoms look and feel as real anatomical objects. Additional visualization of internal anatomical and physiological information and sound generated during the interaction with the phantom yield a multimodal approach that can increase performance, didactic value, and immersion into the virtual environment. Compared to classical approaches, this multimodal display is convenient to use, provides realistic tactile properties, and can be partly adjusted to different, e.g., pathological properties. The interactive phantom is exemplified by a virtual human knee joint that can support orthopedic education, especially for the training of clinical knee joint evaluation. It is suggested that the technical principle can be transferred to many other fields of medical education and training such as obstetrics and dentistry.     

基于联合分布适配的水下声源测距算法研究

水下声源被动测距基于接收数据中声源辐射的声压信号,通过特定方法在空域中搜索声源位置参数,是一个参数估计问题。对于参数估计问题,机器学习方法通常将其转化为分类问题,相比于传统匹配场处理(MFP)具有更准确的估计能力,并且无需先验的声场环境信息。但当训练数据和测试数据的概率密度函数服从不同的分布或者训练数据严重不足时,传统机器学习方法下的分类器预测效果通常较差。因此,该文提出基于联合分布适配(JDA)的水下声源测距算法,该算法使用JDA寻找恰当的变换矩阵进行数据映射,从而减小不同数据域间分布差异,实现源域到目标域的迁移。对经过JDA后数据进行实验的结果表明,JDA可以有效降低在不同时间和不同方位的水声场中获取航迹数据之间的差异,使得基于源域训练的分类器对目标域预测结果的均方根误差(RMSE)和平均绝对误差(MAE)降低了超过30%,从而实现对声源更准确的距离估计。     

A federated semi-supervised learning approach for network traffic classification

The classification of network traffic, which involves classifying and identifying the type of network traffic, is the most fundamental step to network service improvement and modern network management. Classic machine learning and deep learning methods have widely adopted in the field of network traffic classification. However, there are two major challenges in practice. One is the user privacy concern in cross-domain traffic data sharing for the purpose of training a global classification model, and the other is the difficulty to obtain large amount of labeled data for training. In this paper, we propose a novel approach using federated semi-supervised learning for network traffic classification, in which the federated server and clients from different domains work together to train a global classification model. Among them, unlabeled data are used on the client side, and labeled data are used on the server side. The experimental results derived from a public dataset show that the accuracy of the proposed approach can reach 97.81%, and the accuracy gap between the federated learning approach and the centralized training method is minimal.     

数据库样本缺失下的雷达辐射源识别

目前,基于机器学习的雷达辐射源识别技术大多以训练集和测试集同分布为假设,当雷达数据库样本不足导致与信号真实分布存在偏差时,传统的分类方法效果不佳.为此,将迁移学习理论引入识别系统,设计了一种基于结构发现与再平衡的雷达辐射源信号识别方法.通过对数据库和待识别辐射源信号样本进行聚类分析发现数据结构信息,通过重采样处理修正其分布差异.将新采样数据输入支持向量机进行训练并对侦收样本进行识别.仿真实验表明,在新训练样本集上学习的模型对测试集的分类性能有了很大的提升.     

基于蒙特卡罗数据集均衡与鲁棒性增量极限学习机的图像自动标注

针对传统图像标注模型存在着训练时间长、对低频词汇敏感等问题,该文提出了基于蒙特卡罗数据集均衡和鲁棒性增量极限学习机的图像自动标注模型.该模型首先对公共图像库的训练集数据进行图像自动分割,选择分割后相应的种子标注词,并通过提出的基于综合距离的图像特征匹配算法进行自动匹配以形成不同类别的训练集.针对公共数据库中不同标注词的数据规模相差较大,提出了蒙特卡罗数据集均衡算法使得各个标注词间的数据规模大体一致.然后针对单一特征描述存在的不足,提出了多尺度特征融合算法对不同标注词图像进行有效的特征提取.最后针对传统极限学习机存在的隐层节点随机性和输入向量权重一致性的问题,提出了鲁棒性增量极限学习,提高了判别模型的准确性.通过在公共数据集上的实验结果表明：该模型可以在很短时间内实现图像的自动标注,对低频词汇具有较强的鲁棒性,并且在平均召回率、平均准确率、综合值等多项指标上均高于现流行的大多数图像自动标注模型.     

Haptic Rendering of Visual Data for the Visually Impaired

We introduce a force-field haptic rendering method for converting visual data to haptic data. The method includes a novel framework to convert specialized 3D map models into force fields. We generate the final force fields by using structure from motion and implicit surface approximation algorithms. Visually impaired people then can learn to navigate with these maps, using off-the-shelf haptic