Output-relevant Variational autoencoder for Just-in-time soft sensor modeling with missing data

Main challenges for developing data-based models lie in the existence of high-dimensional and possibly missing observations that exist in stored data from industry process. Variational autoencoder (VAE) as one of the deep learning methods has been applied for extracting useful information or features from high-dimensional dataset. Considering that existing VAE is unsupervised, an output-relevant VAE is proposed for extracting output-relevant features in this work. By using correlation between process variables, different weight is correspondingly assigned to each input variable. With symmetric Kullback–Leibler (SKL) divergence, the similarity is evaluated between the stored samples and a query sample. According to the values of the SKL divergence, data relevant for modeling are selected. Subsequently, Gaussian process regression (GPR) is utilized to establish a model between the input and the corresponding output at the query sample. In addition, owing to the common existence of missing data in output data set, the parameters and missing data in the GPR are estimated simultaneously. A practical debutanizer industrial process is utilized to illustrate the effectiveness of the proposed method.     

Mixture ratio modeling of dynamic systems

Any knowledge extraction relies (possibly implicitly) on a hypothesis about the modelled-data dependence. The extracted knowledge ultimately serves to a decision-making (DM). DM always faces uncertainty and this makes probabilistic modelling adequate. The inspected black-box modeling deals with “universal” approximators of the relevant probabilistic model. Finite mixtures with components in the exponential family are often exploited. Their attractiveness stems from their flexibility, the cluster interpretability of components and the existence of algorithms for processing high-dimensional data streams. They are even used in dynamic cases with mutually dependent data records while regression and auto-regression mixture components serve to the dependence modeling. These dynamic models, however, mostly assume data-independent component weights, that is, memoryless transitions between dynamic mixture components. Such mixtures are not universal approximators of dynamic probabilistic models. Formally, this follows from the fact that the set of finite probabilistic mixtures is not closed with respect to the conditioning, which is the key estimation and predictive operation. The paper overcomes this drawback by using ratios of finite mixtures as universally approximating dynamic parametric models. The paper motivates them, elaborates their approximate Bayesian recursive estimation and reveals their application potential.     

Efficient Low Dissipative High Order Schemes for Multiscale MHD Flows, II: Minimization of ∇·B Numerical Error

An adaptive numerical dissipation control in a class of high order filter methods for compressible MHD equations is systematically discussed. The filter schemes consist of a divergence-free preserving high order spatial base scheme with a filter approach which can be divergence-free preserving depending on the type of filter operator being used, the method of applying the filter step, and the type of flow problem to be considered. Some of these filter variants provide a natural and efficient way for the minimization of the divergence of the magnetic field (∇·B) numerical error in the sense that commonly used divergence cleaning is not required. Numerical experiments presented emphasize the performance of the ∇·B numerical error. Many levels of grid refinement and detailed comparison of the filter methods with several commonly used compressible MHD shock-capturing schemes will be illustratedA condensed version appears in the Proceedings of the International Conference on High Performance Scientific Computing, March 10-14, 2003, Hanoi, Vietnam. This is a revised version of a longer internal report, Feb. 19, 2004. The longer internal report was published as a RIACS Technical Report TR03.10, July 2003, NASA Ames Research Center     

激光二极管线阵光束变换研究

推导出激光二极管线阵光束通过透镜系统的光强分布表达式，分析了出射光束的传输特性，设计一种凹柱透镜的光束变换系统得到一维均匀光强分布。此外，给出扩展角宽度的计算公式。     

双轴旋转惯导系统误差传播特性分析

针对双轴旋转惯导系统的误差传播特性问题,推导出双轴旋转惯导系统的误差方程,通过特征值判据证明了误差方程在李雅普诺夫意义下的稳定性,设计了一种合理的十六次序转位方案,建立了研究误差传播特性的方法,分别仿真了初始对准误差角、陀螺常值漂移、陀螺安装误差、陀螺标度因数误差4种因素对导航误差传播特性的影响。结果表明,4种误差因素大小均与导航误差大小成线性关系,并且满足叠加性;4种误差因素会引起导航姿态、导航速度、导航纬度的常值误差和振荡误差,但不会引起积累误差;陀螺标度因数误差会引起导航经度误差的积累和发散。最后提出了减小双轴旋转惯导系统导航误差的措施,对设计双轴旋转惯导系统有一定的参考价值。     

Analysis of Sequence Divergence in Mammalian ABCGs Predicts a Structural Network of Residues That Underlies Functional Divergence

The five members of the mammalian G subfamily of ATP-binding cassette transporters differ greatly in their substrate specificity. Four members of the subfamily are important in lipid transport and the wide substrate specificity of one of the members, ABCG2, is of significance due to its role in multidrug resistance. To explore the origin of substrate selectivity in members 1, 2, 4, 5 and 8 of this subfamily, we have analysed the differences in conservation between members in a multiple sequence alignment of ABCG sequences from mammals. Mapping sets of residues with similar patterns of conservation onto the resolved 3D structure of ABCG2 reveals possible explanations for differences in function, via a connected network of residues from the cytoplasmic to transmembrane domains. In ABCG2, this network of residues may confer extra conformational flexibility, enabling it to transport a wider array of substrates.     

基于全散度的C-V模型阈值法

为了提高传统C-V模型的收敛速度并降低其对噪声的敏感性,提出基于全散度的C-V模型及其快速阈值分割算法。将Bregman散度与全散度统一获得新的全散度,并将其引入C-V模型的拟合偏差项,提高图像灰度值与分割区域平均灰度偏差值计算的鲁棒性。同时,采用变分水平集理论获得基于直方图的快速阈值计算方法。实验结果表明,该方法分割效果及收敛速度得到提高,且具有较好的鲁棒性和抗噪性。     

一种结合主题模型的推荐算法

针对传统协同过滤推荐算法存在的冷启动、数据稀疏以及相似度度量的准确性问题,基于LDA主题模型对文本隐式主题挖掘的有效性和KL散度在主题分布相似性度量的准确性,提出了结合LDA主题模型的矩阵分解推荐算法。首先,利用改进的LDA算法输出项目-主题分布,并用困惑度作为主题数设置的修正函数;然后分别基于余弦相似度和KL散度计算得到项目相似度矩阵,将得到的相似度矩阵结合原评分训练集输出预评分,再将预评分填充到训练集;最后将训练集输入ALS矩阵分解算法得到推荐结果。通过MovieLens数据集的实验结果表明,该算法在不同隐式参数设定下均能得到比ALS推荐算法以及更小的预测误差,并且最优预测误差小于传统推荐算法。该实验说明了通过集成LDA主题模型的ALS算法效果要优于其他推荐算法。     

Towards adaptive learning with improved convergence of deep belief networks on graphics processing units

In this paper we focus on two complementary approaches to significantly decrease pre-training time of a deep belief network (DBN). First, we propose an adaptive step size technique to enhance the convergence of the contrastive divergence (CD) algorithm, thereby reducing the number of epochs to train the restricted Boltzmann machine (RBM) that supports the DBN infrastructure. Second, we present a highly scalable graphics processing unit (GPU) parallel implementation of the CD-k algorithm, which boosts notably the training speed. Additionally, extensive experiments are conducted on the MNIST and the HHreco databases. The results suggest that the maximum useful depth of a DBN is related to the number and quality of the training samples. Moreover, it was found that the lower-level layer plays a fundamental role for building successful DBN models. Furthermore, the results contradict the pre-conceived idea that all the layers should be pre-trained. Finally, it is shown that by incorporating multiple back-propagation (MBP) layers, the DBNs generalization capability is remarkably improved.     

Improved GPU SIMD control flow efficiency via hybrid warp size mechanism

High single instruction multiple data (SIMD) efficiency and low power consumption have made graphic processing units (GPUs) an ideal platform for many complex computational applications. Thousands of threads can be created by programmers and grouped into fixed-size SIMD batches, known as warps. High throughput is then achieved by concurrently executing such warps with minimal control overhead. However, if a branch instruction occurs, which assigns different paths to different threads, one warp will be broken into multiple warps that have to be executed serially, consequently reducing the efficiency advantage of SIMD. In this paper, the contemporary fixed-size warp design is abandoned for a hybrid warp size (HWS) mechanism. Mixed-size warps are generated according to HWS and are scheduled and issued flexibly. The simulation results show that this mechanism yields an average speedup of 1.20 over the baseline architecture for a wide variety of general purpose GPU applications. The paper also integrates HWS with dynamic warp formation (DWF), which is a well-known branch handling mechanism used to improve SIMD utilization by forming new warps out of split warps in real time. The simulation results show that the combination of DWF and HWS generates an average speedup of 1.27 over the DWF-only platform with an estimated area increase of about 1% of DWF.