基于局部深度一致性的自监督手部姿态估计

基于深度图的3D手部姿态估计通常需要大量人工标注数据以达到高精确度和鲁棒性，然而关节点标注过程冗杂且存在一定误差.现有研究工作使用自监督方法解决对标注数据的依赖，通过在虚拟数据集上预训练网络，并在无标注的真实数据集上进行模型拟合，实现3D姿态估计.自监督方法的关键在于设计模型拟合的能量函数以减小模型在真实数据集上的精度下降程度.为了减小模型拟合难度，本文提出局部深度一致性损失，依据初始姿态估计结果，提取输入与输出深度图的局部表征，将深度图显式地解耦为以关节点为中心的不同区域.通过有针对性地对不同关节点进行局部优化，减少虚拟与真实深度图之间的固有领域误差对网络学习的影响，增加训练的稳定性.本文方法在NYU数据集上相比基础方法平均关节点误差提升了21.9%.     

基于差分卷积的自适应视线估计

文章提出了一种基于差分卷积神经网络的自适应视线估计模型。在模型中,融入头部姿态信息,利用差分卷积设计了一种差分网络(Differential Network,DNet),通过训练该网络来预测眼睛的凝视差异,用以校准初步视线估计结果,进而降低视线估计误差。通过在公开数据集Eyediap上进行验证,并与其他性能良好的视线估计模型进行比较,结果均表明所提出的视线估计模型在头部自由运动的状态下可以更准确地估计视线方向。     

基于状态估计的配电网多源信息数据辨识

提出了基于状态估计的配网多源信息数据辨识方法。分析了配电网络等值电路模型以及多源数据量测的状态估计原理。针对非健全信息数据辨识提出了最小二乘估计模型,建立了以N-1故障后误差最小为目标的优化模型,计及了支路潮流约束、节点电压约束、节点注入功率约束,分析了多源数据中的非健全信息量测辨识过程。针对所提出的模型进行了仿真分析,验证了本文所提方法的有效性。     

基于改进堆叠沙漏网络的交警姿态估计

针对交警动作中的姿态估计问题,提出一种改进的堆叠沙漏网络模型。该模型通过减少沙漏网络级联次数,来简化堆叠沙漏网络结构。利用多尺度下深浅层特征信息之间的聚合,得到丰富的上下文信息,增强姿态、遮挡、低分辨率图像的鲁棒性。将不同阶段产生的热图估计结果进行融合平均化处理,进一步提高局部位置坐标的精细定位以及整体估计结果的准确性。在MPII数据集以及中国交警数据集上进行实验,结果表明,改进后的网络模型提高了运行的效率,同时可以很好地对目标交警的姿态特征信息进行提取,对结果热图平均化处理后,提高了位置坐标整体估计的准确性。     

云计算环境下的网络安全估计模型态势仿真

在云计算环境下,传统方法采用终端网络监测方法进行网络安全估计,由于网络通信信道终端功率衰减性强,导致安全态势估计精度不高,检测性能不好。提出一种基于自适应数据分类和病毒感染隶属度特征提取的云计算环境下网络安全估计及态势预测算法。构建云计算环境下的网络安全估计模型,采用自适应数据分类算法对网络攻击信息数据进行聚类评估,提取网络攻击病毒数据的感染隶属度特征,实现网络安全态势预测和病毒攻击检测。仿真实验表明,该算法对病毒数据流预测精度较高,实现不同场景下的网络病毒流预测和数据检测,提高了云计算环境下网络抵御病毒攻击的能力。     

大数据环境下的Web网络风险估计

针对传统大数据评估过程中的Web网络风险评估结果不精准并且速度较慢的问题,提出一种大数据环境下的Web网络风险估计方法。采用特定的风险评估因子进行有效的评估,避免了传统方法中评定项目繁杂、计算量大等因素造成的评估不准的问题。评估过程中使用了特定的评估模型,把传统的加权平均的风险估计方法转变成为非线性映射评估方法,这样能够更加准确的进行风险评估。为了验证设计的大数据环境下的Web网络风险估计的有效性,设计了对比仿真实验。实验结果表明,设计的大数据环境下的Web网络风险估计方法能够有效地解决风险评估过程中的评估结果不准确问题。     

稀疏孔径和大转角下ISAR对目标转动的估计

逆合成孔径雷达（inverse synthetic aperture radar，ISAR）对非合作目标做成像时图像质量依赖于对目标运动参数的准确估计.针对在稀疏孔径和非均匀转动条件下现存的参数估计方法计算量过大或者方法适用条件不满足，提出了一种基于神经网络的参数估计方法.此方法以成像问题的模型知识指导数据的生成过程，然后训练通用的神经网络，最终实现将数据中隐含的知识转化为转动估计器.从仿真实验结果来看，所得到的网络对满足一定信噪比要求的回波数据可以提供较准确的估计，所得参数可以帮助成像算法提高聚焦效果，大量的样例表明网络可以部分学习到回波与转动之间的关系.     

基于1D-Concatenate的信道估计DNN模型优化方法

为提高DNN模型在无线通信中信道估计精度,提出一种基于1D-Concatenate的信道估计DNN模型优化方法。该方法将Concatenate进行一维（1D）数据转换,以跳跃连接的方式引入DNN模型,抑制梯度消失问题,运用1D-Concatenate恢复网络训练过程中丢失的数据特征,提高DNN信道估计精度。为验证优化方法的有效性,选取较典型的基于DNN的无线通信信道估计模型进行对比仿真实验。实验结果表明,本文提出的优化方法对已有DNN模型的估计增益提升可达77.10%,在高信噪比下信道增益提升可达3 dB。该优化方法能有效提高DNN模型在无线通信中的信道估计精度,特别是高信噪比下提升效果显著。     

基于射线跟踪的小区信号强度估计

基于经验模型的无线信号强度估计方法无法针对具体物理场景,估计结果精度低,无法满足移动通信网络规划和优化的需要。射线跟踪技术能依据场景信息跟踪电波传播路径,但现有的反向跟踪方法在进行室外成片区域信号强度估计时复杂度高,无法实用。为提高估计结果的精度,降低估计方法的复杂度,提出了一种正向跟踪信号强度估计方法,将基站天线处发射的电磁波模拟成离散的射线,在考虑建筑物形状、分布信息的基础上采用基于网格的搜索算法跟踪射线路径,在接收点处对反射、绕射射线接收合并,最终得到基站覆盖区域的信号强度分布。仿真结果表明该方法的场强估计结果和实测数据吻合程度远大于经验模型估计方法,并且80%左右的估计结果和实测数据的误差在12 dBm以内,可应用于移动通信网络规划和优化。     

基于k阶马尔可夫链的单播网络丢包层析成像

该文针对时域相关的网络链路丢包估计问题,提出一种基于k阶马尔可夫链的单播网络丢包层析成像方法。该方法首先引入k阶马尔可夫链描述网络链路丢包过程,然后用最大伪似然方法估计k阶马尔可夫链链路丢包模型的状态转移概率。当k足够大时,该文方法可以根据单播端到端测量数据,准确地估计出网络链路上每个数据包丢失的概率。ns-2仿真验证了该文方法的有效性。     

Removing label ambiguity in learning-based visual saliency estimation

Visual saliency is a useful clue to depict visually important image/video contents in many multimedia applications. In visual saliency estimation, a feasible solution is to learn a "feature-saliency" mapping model from the user data obtained by manually labeling activities or eye-tracking devices. However, label ambiguities may also arise due to the inaccurate and inadequate user data. To process the noisy training data, we propose a multi-instance learning to rank approach for visual saliency estimation. In our approach, the correlations between various image patches are incorporated into an ordinal regression framework. By iteratively refining a ranking model and relabeling the image patches with respect to their mutual correlations, the label ambiguities can be effectively removed from the training data. Consequently, visual saliency can be effectively estimated by the ranking model, which can pop out real targets and suppress real distractors. Extensive experiments on two public image data sets show that our approach outperforms 11 state-of-the-art methods remarkably in visual saliency estimation.     

Registration-based compensation using sparse representation in conformal-array STAP

Space-time adaptive processing (STAP) is a well-known technique in detecting slow-moving targets in the clutter-spreading environment. When considering the STAP system with conformal radar array (CFA), the training data are range-dependent, which results in poor detection performance of traditional statistical-based algorithms. Current registration-based compensation (RBC) is implemented based on a sub-snapshot spectrum using temporal smoothing. In this case, the estimation accuracy of the configuration parameters and the clutter power distribution is limited. In this paper, the technique of sparse representation is introduced into the spectral estimation, and a new compensation method is proposed, namely RBC with sparse representation (SR-RBC). This method first establishes the relationship between the clutter covariance matrix (CCM) and the clutter spectral distribution. Based on this, it avoids the problem of lacking stationary training data and converts the CCM estimation into the solving of the underdetermined equation only with the test cell. Then sparse representation method, like iterative reweighted least square (IRLS) is used to guide the solution of the underdetermined equation towards the actual clutter distribution. Finally, the transform matrix is designed using the CCM estimation so that the processed training data behaves nearly stationary with the test cell. Because the configuration parameters and the clutter spectral response are obtained with full-snapshot using sparse representation, SR-RBC provides more accurate clutter spectral estimation, and the transformed training data are more stationary so that better signal-clutter-ratio (SCR) improvement is achieved.     

Quantum algorithm for soft margin support vector machine withhinge loss function

Soft margin support vector machine (SVM) with hinge loss function is an important classification algorithm, which has been widely used in image recognition, text classification and so on. However, solving soft margin SVM with hinge loss function generally entails the sub-gradient projection algorithm, which is very time-consuming when processing big training data set. To achieve it, an efficient quantum algorithm is proposed. Specifically, this algorithm implements the key task of the sub-gradient projection algorithm to obtain the classical sub-gradients in each iteration, which is mainly based on quantum amplitude estimation and amplification algorithm and the controlled rotation operator. Compared with its classical counterpart, this algorithm has a quadratic speedup on the number of training data points. It is worth emphasizing that the optimal model parameters obtained by this algorithm are in the classical form rather than in the quantum state form. This enables the algorithm to classify new data at little cost when the optimal model parameters are determined.     

On adaptive decision rules and decision parameter adaptation forautomatic speech recognition

Recent advances in automatic speech recognition are accomplished by designing a plug-in maximum a posteriori decision rule such that the forms of the acoustic and language model distributions are specified and the parameters of the assumed distributions are estimated from a collection of speech and language training corpora. Maximum-likelihood point estimation is by far the most prevailing training method. However, due to the problems of unknown speech distributions, sparse training data, high spectral and temporal variabilities in speech, and possible mismatch between training and testing conditions, a dynamic training strategy is needed. To cope with the changing speakers and speaking conditions in real operational conditions for high-performance speech recognition, such paradigms incorporate a small amount of speaker and environment specific adaptation data into the training process. Bayesian adaptive learning is an optimal way to combine prior knowledge in an existing collection of general models with a new set of condition-specific adaptation data. In this paper, the mathematical framework for Bayesian adaptation of acoustic and language model parameters is first described. Maximum a posteriori point estimation is then developed for hidden Markov models and a number of useful parameters densities commonly used in automatic speech recognition and natural language processing     

Channel estimation using implicit training

In this paper, a new method to perform channel estimation is presented. It is shown that accurate estimation can be obtained when a training sequence is actually arithmetically added to the information data as opposed to being placed in a separate empty time slot: hence, the word "implicit." A closed-form solution for the estimation variance is derived, as well as the Cramer-Rao lower bound. Conditions are derived for the training sequences that result in a channel estimation performance that is independent of the channel characteristics. In addition, estimation performance is shown to be independent of the modulation format. A procedure to synthesize optimal training sequences is presented, and the problem of synchronization is solved. The performance of the algorithm is then compared with other methods that use explicit training under GSM-like environmental conditions, and the new algorithm is shown to be competitive with these. Finally, comparisons are also carried out against blind methods over realistic bandlimited channels, and these show that the new method exhibits good performance.     

Doubly-Selective Channel Estimation Using Superimposed Training and Weighted First-Order Statistics

Doubly-selective channel estimation using superimposed training and complex exponential basis expansion model is considered. By taking a weighted averaging operation of the received data, a weighted first-order statistical estimator is proposed, where the time-varying channel estimation is reduced to the simple average-based solution of time-invariant coefficients and the dominant effect of information-induced interference on channel estimation can be suppressed. To further improve the estimation performance with a limited training power, a joint iterative channel estimation and symbol detection scheme is developed where the detected symbol is exploited to enhance estimation performance instead of being viewed as interference. Theoretical analysis and simulation results show that the proposed scheme is superior to data-dependent superimposed training scheme and competitive with the conventional time-multiplexed training in terms of symbol error rate over doubly-selective channels.     

Orthogonal superimposed training design for block transmission over frequency‐selective fading channels

Channel estimation for single‐carrier block transmission over frequency‐selective fading channel using superimposed training is addressed. A novel affine precoding model based on orthogonal polyphase sequence set is designed to decouple channel estimation from symbol detection. The orthogonal constraints on the training and precoding matrices ensure the separation of superimposed signals and accurate channel estimation with less training overheads as compared with time‐multiplexed scheme. Simulation results show that the proposed scheme exhibits good performance and outperforms another data‐dependent superimposed training scheme, especially for compact constellations or channel with long delay‐spread. Copyright © 2014 John Wiley & Sons, Ltd.     

A PMC-driven methodology for energy estimation in RVC-CAL video codec specifications

In this paper, an energy estimation methodology based on performance monitor counters (PMC) is proposed to estimate the energy consumption of RVC-CAL video codec specifications. The proposed PMC-driven methodology is able to automatically identify the most appropriate events and training data to cover the main application characteristics. In addition, knowledge of the hardware platform employed is not required. Therefore, this methodology can be easily implemented on other PMC-available systems while keeping the estimation accuracy. It is worth noting that this is an attractive asset to analyze the energy consumption of RVC-CAL codec specifications. Besides, the methodology reduces the PMC redundancy and, thus, the overhead introduced when applied to on-line power management. Experimenting on two RVC-CAL decoders, H.264 and MPEG4 Part2 SP, a coarse estimation model based on instructions per cycle (IPC) and the proposed PMC-driven model are compared. The results show that the PMC-driven model can achieve for the H.264 and MPEG4 Part2 SP decoders average estimation errors of 5.95% and 5.01%, respectively, in comparison to the 17.11% and 13.65% average errors obtained with the IPC-based model. As a consequence, this methodology is suggested to be combined into the RVC framework to help the designer to have an overview of the energy consumption of the specification actors at earlier design stages.     

Carrier Frequency Offset Estimation using Data-Dependent Superimposed Training

In this letter we propose, for the first time, a solution to the problem of carrier frequency offset (CFO) estimation within the data dependent superimposed training (DDST) framework for channel estimation. While time division multiplexed (TDM) trained systems can use the TDM sequence to determine the CFO, the original attraction of DDST for channel estimation was that it avoided any TDM training. So in this letter we show how CFO estimation can still be very effectively performed with the DDST algorithm, while continuing to preclude the need for any additional bandwidth-consuming TDM training. Finally, simulations are presented that verify the theoretical results.