Tree-Structured Stick Breaking Processes for Hierarchical Data

Many data are naturally modeled by an unobserved hierarchical structure. In this paper we propose a flexible nonparametric prior over unknown data hierarchies. The approach uses nested stick-breaking processes to allow for trees of unbounded width and depth, where data can live at any node and are infinitely exchangeable. One can view our model as providing infinite mixtures where the components have a dependency structure corresponding to an evolutionary diffusion down a tree. By using a stick-breaking approach, we can apply Markov chain Monte Carlo methods based on slice sampling to perform Bayesian inference and simulate from the posterior distribution on trees. We apply our method to hierarchical clustering of images and topic modeling of text data.

Coupled foundation-superstructure analysis and influence of building stiffness on foundation response

This paper proposes a simple method to include superstructure stiffness in foundation analyses. The method involves extracting a small "condensed structural matrix" from finite element models of the superstructure, which can then be incorporated into pile group or piled raft analyses using common approaches such as elastic continuum or load transfer methods. The matrix condensation method directly couples structural and geotechnical analyses, and eliminates the need for iterative analyses between structural and geotechnical engineers. Effectiveness of the approach is illustrated through analyses of several buildings designed with a typical floor plan but with varying heights. The parametric study illustrates that superstructure stiffness can have a significant influence on foundation settlement estimates, and the stiffening effects are dominated by the lower stories of the superstructure. The proposed method aims to bridge the gap between structural and geotechnical analyses. Also, being a computationally simple and accurate approach, it is applicable to parametric or optimization studies that would otherwise involve large amounts of analyses. © 2010 ASCE.

Simulated Annealing: Rigorous finite-time guarantees for optimization on continuous domains

Simulated annealing is a popular method for approaching the solution of a global optimization problem. Existing results on its performance apply to discrete combinatorial optimization where the optimization variables can assume only a finite set of possible values. We introduce a new general formulation of simulated annealing which allows one to guarantee finite-time performance in the optimization of functions of continuous variables. The results hold universally for any optimization problem on a bounded domain and establish a connection between simulated annealing and up-to-date theory of convergence of Markov chain Monte Carlo methods on continuous domains. This work is inspired by the concept of finite-time learning with known accuracy and confidence developed in statistical learning theory.

A reversible infinite HMM using normalised random measures

Copyright 2014 by the author(s). We present a nonparametric prior over reversible Markov chains. We use completely random measures, specifically gamma processes, to construct a countably infinite graph with weighted edges. By enforcing symmetry to make the edges undirected we define a prior over random walks on graphs that results in a reversible Markov chain. The resulting prior over infinite transition matrices is closely related to the hierarchical Dirichlet process but enforces reversibility. A reinforcement scheme has recently been proposed with similar properties, but the de Finetti measure is not well characterised. We take the alternative approach of explicitly constructing the mixing measure, which allows more straightforward and efficient inference at the cost of no longer having a closed form predictive distribution. We use our process to construct a reversible infinite HMM which we apply to two real datasets, one from epigenomics and one ion channel recording.

SpMV的自动性能优化实现技术及其应用研究

在科学计算中,稀疏矩阵向量乘(SpMV)是一个十分重要且经常被大量调用的计算内核.由于SpMV一般实现算法的浮点计算和存储访问次数比率非常低,且其存储访问模式极为不规则,其实际运行性能往往很低.通过采用寄存器分块算法和启发式分块大小选择算法,将稀疏矩阵分成小的稠密分块,重用保存在寄存器中向量x元素,可以提高该计算内核的性能.剖析和总结了OSKI软件包所采用的若干关键优化技术,并进行了实际应用性能测试.测试表明,在实际应用这些优化技术的过程中,应用程序对SpMV的调用次数要达到上百次的量级,才能抵消由于应用这些性能优化技术所带来的额外时间开销,取得性能加速效果.在Pentium4和AMD Athlon平台上,测试了10个矩阵,其平均加速比分别达到了1.69和1.48.

RAM(h)模型下SpMV存储访问复杂度的分析

稀疏矩阵向量乘(SpMV)采取压缩行存储格式的算法性能非常差,而寄存器分块算法可以使得数据尽量在靠近处理器的存储层次中访问而提高性能.利用RAM(h)模型进行分析和比较不同算法形式的存储访问复杂度,可以比较两种算法的优劣.通过RAM(h)分析SpMV两种实现形式的存储访问复杂度,同时在奔腾四平台上,测试了7个稀疏矩阵的SpMV性能,并统计了这两种算法中L1,L2,和TLB的缺失率,实验结果与模型分析的数据一致.

OpenMP循环调度算法及SpMV多核并行化研究

OpenMP是一种支持Fortran，C/C++的共享存储并行编程标准。它基于fork-join的并行执行模型，将程序划分为并行区和串行区。近几年来，OpenMP在SMP(Symmetric Multi-Processing)和多核体系结构的并行编程中得到了广泛的应用。随着多核处理器的发展，实际的应用程序如何充分利用多个处理器核来提高运算效率也成为研究的热点。 在科学计算中，循环结构是最核心的并行对象之一。考虑到负载平衡、调度开销、同步开销等多方面因素，OpenMP标准制定了Static调度、Dynamic调度、Guided调度和Runtime调度等不同策略。针对Guided调度策略不适合递减型循环结构的缺点，本文提出了一种改进的new_guided调度策略，并在OMPi编译器上加以实现。New_guided调度策略的主要思想是对前半部分的循环采用Static调度，后半部分的循环采用Guided调度。此外，本文针对不同的循环结构，在多核处理器上对不同的调度策略进行了评测。测试结果表明，在一般情况下，OpenMP默认的Static策略的调度性能最差；对于规则的循环结构和递增的循环结构，Dynamic调度策略、Guided调度策略和new_guided策略的性能差别不大；对于递减型的循环结构，Dynamic调度策略和new_guided策略的性能相当，要优于Guided调度策略；对于求解Mandelbrot集合这类计算量集中在中间的随机循环结构，Dynamic调度策略优于其它策略，new_guided策略的性能介于Dynamic调度和Guided调度之间。 随着多核处理器的问世和发展，多线程程序设计也已经成为一个不可回避的问题。稀疏矩阵向量乘(SpMV, Sparse Matrix-Vector Multiplication)是一个十分重要且经常被大量调用的科学计算内核。SpMV的存储访问一般都极不规则，导致现有的SpMV算法效率都比较低。目前，多核处理器芯片上的内核数量正在逐步增加。这使得在多核处理器上对SpMV进行并行化加速变得非常重要。本文介绍了稀疏矩阵的两种常用的存储格式CSR和BCSR，并采用OpenMP实现了SpMV的多核并行化。此外，本文还讨论了寄存器分块算法、压缩列索引等优化技术，以及不同调度策略对多线程并行后的SpMV的影响。在曙光天阔服务器S4800A1上的测试表明，大部分矩阵都取得了可扩展、甚至是超线性的加速比，但是对于部分规模较大的矩阵，加速效果并不明显。在我们的测试中，与基于CSR实现的多线程SpMV相比，采用寄存器分块算法优化后的SpMV运算速度平均提高了28.09%。在基于CSR实现的多线程SpMV中，采用列索引优化技术后的程序比优化前的速度平均提高了13.05%。此外，本文实现了一种基于非零元个数的调度策略。在该策略中，每个线程处理几乎相同数量的非零元。我们将它和OpenMP标准提供的三种调度策略进行了测试和分析。测试结果表明：与OpenMP提供的调度策略相比，基于非零元个数的调度策略能取得更好的负载平衡；Dynamic调度和Guided调度在多线程SpMV中的性能基本相当，均优于Static调度策略。

地震波场模拟及叠前深度偏移的李群方法

The processes of seismic wave propagation in phase space and one way wave extrapolation in frequency-space domain, if without dissipation, are essentially transformation under the action of one parameter Lie groups. Consequently, the numerical calculation methods of the propagation ought to be Lie group transformation too, which is known as Lie group method. After a fruitful study on the fast methods in matrix inversion, some of the Lie group methods in seismic numerical modeling and depth migration are presented here. Firstly the Lie group description and method of seismic wave propagation in phase space is proposed, which is, in other words, symplectic group description and method for seismic wave propagation, since symplectic group is a Lie subgroup and symplectic method is a special Lie group method. Under the frame of Hamiltonian, the propagation of seismic wave is a symplectic group transformation with one parameter and consequently, the numerical calculation methods of the propagation ought to be symplectic method. After discrete the wave field in time and phase space, many explicit, implicit and leap-frog symplectic schemes are deduced for numerical modeling. Compared to symplectic schemes, Finite difference (FD) method is an approximate of symplectic method. Consequently, explicit, implicit and leap-frog symplectic schemes and FD method are applied in the same conditions to get a wave field in constant velocity model, a synthetic model and Marmousi model. The result illustrates the potential power of the symplectic methods. As an application, symplectic method is employed to give synthetic seismic record of Qinghai foothills model. Another application is the development of Ray+symplectic reverse-time migration method. To make a reasonable balance between the computational efficiency and accuracy, we combine the multi-valued wave field & Green function algorithm with symplectic reverse time migration and thus develop a new ray+wave equation prestack depth migration method. Marmousi model data and Qinghai foothills model data are processed here. The result shows that our method is a better alternative to ray migration for complex structure imaging. Similarly, the extrapolation of one way wave in frequency-space domain is a Lie group transformation with one parameter Z and consequently, the numerical calculation methods of the extrapolation ought to be Lie group methods. After discrete the wave field in depth and space, the Lie group transformation has the form of matrix exponential and each approximation of it gives a Lie group algorithm. Though Pade symmetrical series approximation of matrix exponential gives a extrapolation method which is traditionally regarded as implicit FD migration, it benefits the theoretic and applying study of seismic imaging for it represent the depth extrapolation and migration method in a entirely different way. While, the technique of coordinates of second kind for the approximation of the matrix exponential begins a new way to develop migration operator. The inversion of matrix plays a vital role in the numerical migration method given by Pade symmetrical series approximation. The matrix has a Toepelitz structure with a helical boundary condition and is easy to inverse with LU decomposition. A efficient LU decomposition method is spectral factorization. That is, after the minimum phase correlative function of each array of matrix had be given by a spectral factorization method, all of the functions are arranged in a position according to its former location to get a lower triangular matrix. The major merit of LU decomposition with spectral factorization (SF Decomposition) is its efficiency in dealing with a large number of matrixes. After the setup of a table of the spectral factorization results of each array of matrix, the SF decomposition can give the lower triangular matrix by reading the table. However, the relationship among arrays is ignored in this method, which brings errors in decomposition method. Especially for numerical calculation in complex model, the errors is fatal. Direct elimination method can give the exact LU decomposition But even it is simplified in our case, the large number of decomposition cost unendurable computer time. A hybrid method is proposed here, which combines spectral factorization with direct elimination. Its decomposition errors is 10 times little than that of spectral factorization, and its decomposition speed is quite faster than that of direct elimination, especially in dealing with a large number of matrix. With the hybrid method, the 3D implicit migration can be expected to apply on real seismic data. Finally, the impulse response of 3D implicit migration operator is presented.

Modeling the evolution of regulatory elements by simultaneous detection and alignment with phylogenetic pair HMMs.

The computational detection of regulatory elements in DNA is a difficult but important problem impacting our progress in understanding the complex nature of eukaryotic gene regulation. Attempts to utilize cross-species conservation for this task have been hampered both by evolutionary changes of functional sites and poor performance of general-purpose alignment programs when applied to non-coding sequence. We describe a new and flexible framework for modeling binding site evolution in multiple related genomes, based on phylogenetic pair hidden Markov models which explicitly model the gain and loss of binding sites along a phylogeny. We demonstrate the value of this framework for both the alignment of regulatory regions and the inference of precise binding-site locations within those regions. As the underlying formalism is a stochastic, generative model, it can also be used to simulate the evolution of regulatory elements. Our implementation is scalable in terms of numbers of species and sequence lengths and can produce alignments and binding-site predictions with accuracy rivaling or exceeding current systems that specialize in only alignment or only binding-site prediction. We demonstrate the validity and power of various model components on extensive simulations of realistic sequence data and apply a specific model to study Drosophila enhancers in as many as ten related genomes and in the presence of gain and loss of binding sites. Different models and modeling assumptions can be easily specified, thus providing an invaluable tool for the exploration of biological hypotheses that can drive improvements in our understanding of the mechanisms and evolution of gene regulation.

Applications of Feller-Reuter-Riley transition functions

A Feller–Reuter–Riley function is a Markov transition function whose corresponding semigroup maps the set of the real-valued continuous functions vanishing at infinity into itself. The aim of this paper is to investigate applications of such functions in the dual problem, Markov branching processes, and the Williams-matrix. The remarkable property of a Feller–Reuter–Riley function is that it is a Feller minimal transition function with a stable q-matrix. By using this property we are able to prove that, in the theory of branching processes, the branching property is equivalent to the requirement that the corresponding transition function satisfies the Kolmogorov forward equations associated with a stable q-matrix. It follows that the probabilistic definition and the analytic definition for Markov branching processes are actually equivalent. Also, by using this property, together with the Resolvent Decomposition Theorem, a simple analytical proof of the Williams' existence theorem with respect to the Williams-matrix is obtained. The close link between the dual problem and the Feller–Reuter–Riley transition functions is revealed. It enables us to prove that a dual transition function must satisfy the Kolmogorov forward equations. A necessary and sufficient condition for a dual transition function satisfying the Kolmogorov backward equations is also provided.

Population Mobility Dynamics Estimated from Mobile Telephony Data

In the last decade, mobile phones and mobile devices using mobile cellular telecommunication network connections have become ubiquitous. In several developed countries, the penetration of such devices has surpassed 100 percent. They facilitate communication and access to large quantities of data without the requirement of a fixed location or connection. Assuming mobile phones usually are in close proximity with the user, their cellular activities and locations are indicative of the user's activities and movements. As such, those cellular devices may be considered as a large scale distributed human activity sensing platform. This paper uses mobile operator telephony data to visualize the regional flows of people across the Republic of Ireland. In addition, the use of modified Markov chains for the ranking of significant regions of interest to mobile subscribers is investigated. Methodology is then presented which demonstrates how the ranking of significant regions of interest may be used to estimate national population, results of which are found to have strong correlation with census data.

Three-dimensional modeling of the human fallopian tube fimbriae

OBJECTIVE: Ovarian cancer is the most lethal gynecological malignancy that affects women. Recent data suggests that the disease may originate in the fallopian fimbriae; however, the anatomical origin of ovarian carcinogenesis remains unclear. This is largely driven by our lack of knowledge regarding the structure and function of normal fimbriae and the relative paucity of models that accurately recapitulate the in vivo fallopian tube. Therefore, a human three-dimensional (3D) culture system was developed to examine the role of the fallopian fimbriae in serous tumorigenesis.

METHODS: Alginate matrix was utilized to support human fallopian fimbriae ex vivo. Fimbriae were cultured with factors hypothesized to contribute to carcinogenesis, namely; H2O2 (1mM) a mimetic of oxidative stress, insulin (5μg/ml) to stimulate glycolysis, and estradiol (E2, 10nM) which peaks before ovulation. Cultures were evaluated for changes in proliferation and p53 expression, criteria utilized to identify potential precursor lesions. Further, secretory factors were assessed after treatment with E2 to identify if steroid signaling induces a pro-tumorigenic microenvironment.

RESULTS: 3D fimbriae cultures maintained normal tissue architecture up to 7days, retaining both epithelial subtypes. Treatment of cultures with H2O2 or insulin significantly induced proliferation. However, p53 stabilization was unaffected by any particular treatment, although it was induced by ex vivo culturing. Moreover, E2-alone treatment significantly induced its canonical target PR and expression of IL8, a factor linked to poor outcome.

CONCLUSIONS: 3D alginate cultures of human fallopian fimbriae provide an important microphysiological model, which can be further utilized to investigate serous tumorigenesis originating from the fallopian tube.

The effect of redundancy and neutrality in genetic search

An experimental study aimed at assessing the influence of redundancy and neutrality on the performance of an (1+1)-ES evolution strategy modeled using Markov chains and applied to NK fitness landscapes is presented. For the study, two families of redundant binary representations, one non-neutral family which is based on linear transformations and that allows the phenotypic neighborhoods to be designed in a simple and effective way, and the neutral family based on the mathematical formulation of error control codes are used. The results indicate whether redundancy or neutrality affects more strongly the behavior of the algorithm used.