Artificial Intelligence research at the National University of Singapore

This paper reports three on-going research projects in Artificial Intelligence (AI) at the National University of Singapore. These three projects are (a) designing an expert system for computer performance monitoring, (b) machine translation projects involving three common languages in Singapore and (c) neural network systems providing an alternative to conventional computing to solve AI problems.     

TMLNNs：三值／多值逻辑神经元网络

本文提出了具有三值／多值逻辑表达能力的神经元模型,即三值／多值“逻辑与”神经元和三值／多值“逻辑或”神经元,由这种逻辑神经元连接而成的多层神经网络能够实现三值／多值逻辑推理系统。本文还给出了ＴＭＬＮＮｓ的学习算法,从ＴＭＬＮＮｓ网络中容易抽取到三值／多值逻辑规则知识,可以将ＴＭＬＮＮｓ用于三值／多值逻辑规则知识的自动获取,ＴＭＬＮＮｓ模型为神经网络表示逻辑知识提供了理论基础。     

NMF初始化研究及其在文本分类中的应用

对非负矩阵分解的初始化进行研究，提出针对文本分类的主成分分析(PCA)、有监督PCA(SPCA)和模糊C平均3种初始化方法并进行了实验。多类文本分类的实验结果表明，这些方法有效地解决了初值对结果的影响问题，不同程度地提高了文本分类结果，其中SPCA优于其他2种方法。     

Linear cryptanalysis of NUSH block cipher

NUSH is a block cipher as a candidate for NESSIE. NUSH is analyzed by linear crypt-analysis . The complexity δ = (ε , η) of the attack consists of data complexity ε and time complexity η. Three linear approximations are used to analyze NUSH with 64-bit block. When |K| = 128 bits, the complexities of three attacks are (258, 2124), (260, 278) and (262, 255) respectively. When |K| = 192 bits, the complexities of three attacks are (258, 2157) (260, 2%) and (262, 258) respectively. When |K| = 256 bits, the complexities of three attacks are (258, 2125), (260, 278) and (262, 253) respectively. Three linear approximations are used to analyze NUSH with 128-bit block. When |K|= 128 bits, the complexities of three attacks are (2122, 295), (2124, 257) and (2126, 252) respectively. When |K| = 192 bits, the complexities of three attacks are (2122, 2142), (2124, 275) and (2126, 258) respectively. When |K|= 256 bits, the complexities of three attacks are (2122, 2168), (2124, 281) and (2126, 264) respectively. Two l     

语义角色的精细等级及其在信息处理中的应用

本文首先讨论语义角色的三种精细程度不一的分类层级,介绍它们各自在语言信息处理系统中的有关应用。接着,分别介绍三种为语言信息处理服务的语义资源对于语义角色的不同处理: (i)加州大学伯克利分校框架网的语义角色——基于场景的语义框架中的框架元素; (ii)宾州大学命题库的语义角色——基于特定动词的编了号的原型角色; (iii)北京大学中文网库的语义角色——基于特定谓词的各论元成分的论旨角色。最后,从建库目标、方法论、标注内容和系统构成等方面,比较这三个语义关系标注语料库的同异。     

Visualizing the sequential process of knowledge elaboration in computer-supported collaborative problem solving

This case study illustrates the sequential process of the joint and individual knowledge elaboration in a computer-supported collaborative learning (CSCL) environment. The case comprised an Internet-based physics problem-solving platform. Six Dutch secondary school students (three males, three females) participated in the three-week experiment. They were paired based on self-selection. Each dyad was asked to collaborate on eight moderately structured problems concerning Newtonian mechanics. Their online interactions, including their textual and pictorial messages, were categorized and sequentially plotted. The three dyads showed three different collaboration patterns in terms of joint and individual knowledge elaboration.     

语言与图灵测试

讨论了语言的三指功能(指名、指物、指心)与图灵测试的关系,指出早期的人工智能研究一般假设语言三指等价,因此致力于实现其中之一的功能.如果一个功能实现了,其他两个功能也就实现了, 比如图灵测试将智能的表现限定在指名功能里.但是,现代科学已经证明,语言三指不一定等价.这给人工智能的研究带来很大挑战,本文罗列了一些开放性问题,如对话的理论计算机制等.     

Three improved hybrid metaheuristic algorithms for engineering design optimization

This paper presents three hybrid metaheuristic algorithms that further improve the two hybrid differential evolution (DE) metaheuristic algorithms described in Liao [1]. The three improved algorithms are: (i) MDE′–HJ, which is a modification of MA–MDE′ in Liao [1] by replacing the random walk with direction exploitation local search with the Hooke and Jeeves (HJ) method; (ii) MDE′–IHS–HJ, which is constructed by adding the Hooke and Jeeves method to the original cooperative hybrid, i.e., MDE′–IHS; and (iii) PSO–MDE′–HJ, which is a variation of MDE′–IHS–HJ by replacing improved harmony search (IHS) with particle search optimization (PSO). A comprehensive comparative study was carried out to compare the three improved hybrids with the three algorithms presented by Liao [1] in terms of average success rate, average function evaluations taken, average elapsed CPU time, and convergence profiles. A total of 18 problems, 4 more than those used in Liao [1], were selected from different engineering domains for testing. The test results indicate that all three new hybrids can achieve higher success rate in much less CPU time. Among these three hybrids, MDE′–IHS–HJ is the best one in terms of success rate, better than the best hybrid in Liao [1] by over 15% and better than the second best, PSO–MDE′–HJ, by nearly 10%.     

A sensitivity analysis of simulated River Basin planning for capital budgeting decisions

The simulation model in this study is applied to the Grande Ronde River Basin which is located in northeastern Oregon. The exogenous variables in this study include: (1) hydrology data as based on 41 years of historical hydrology (uncontrollable exogenous variable); (2) capital sets as determined for three different sizes of two reservoirs or nine combinations (partially controllable exogenous variable); and (3) operating procedures as determined for three scales of irrigation development (partially controllable exogenous variable). The functional relationships for benefits entail relationships between hydrologie phenomena and benefits for seven uses of water. These seven uses include: (1) irrigation; (2) municipal and industrial water supply; (3) recreation; (4) salmon reared in reservoir; (5) resident trout in reservoir; (6) flood control; and (7) anadromous fish. The endogenous variables resulting from running the model include net present benefits and benefit-cost ratios for nine reservoir combinations and three operating procedures (a decision set of 27). The results from the runs indicate that some decision sets are clearly superior to others and reduces the choice set from 27 to three.     

Modelling and recognition of the linguistic components in American Sign Language

The manual signs in sign languages are generated and interpreted using three basic building blocks: handshape, motion, and place of articulation. When combined, these three components (together with palm orientation) uniquely determine the meaning of the manual sign. This means that the use of pattern recognition techniques that only employ a subset of these components is inappropriate for interpreting the sign or to build automatic recognizers of the language. In this paper, we define an algorithm to model these three basic components form a single video sequence of two-dimensional pictures of a sign. Recognition of these three components are then combined to determine the class of the signs in the videos. Experiments are performed on a database of (isolated) American Sign Language (ASL) signs. The results demonstrate that, using semi-automatic detection, all three components can be reliably recovered from two-dimensional video sequences, allowing for an accurate representation and recognition of the signs.     

Programming three parallel computers

Three commercial systems are considered from a programmer's point of view. The three are the Intel iPSC, a network of Inmos transputers, and the Sequent Balance. The differences in overhead are examined by implementing a solution to the traveling-salesman problem on all three. The evaluation focuses on three major issues in parallel programming: (1) how execution is divided among processing elements and how it is controlled; (2) how data are shared; and (3) how events are synchronized. The experiences of the authors are presented and some specific as well as general conclusions are drawn     

事件检测支持向量机模型与神经网络模型比较

针对交通领域中的事件检测(无事件模式和有事件模式)模式识别问题,描述了支持向量机(SVM)的基本方法,建立了基于线性(linearfunction)、多项式(polynomialfunction)和径向基(radialbasisfunction)3种核函数的事件检测SVM模型,并与PNN、MLF模型进行了理论比较。采用I-880线圈数据集和事件数据集建立并验证SVM、PNN和MLF模型,结果发现:无论对于向北、向南或混合方向的事件检测,SVM模型的检测率(DR)、误报率(FAR)和平均检测时间(MTTD)指标均比MLF模型好;PNN模型的DR比SVM(P)模型的高,但FAR和MTTD指标不比SVM(P)模型好;在3个SVM模型中,SVM(P)检测效果最好,SVM(L)最差。SVM算法与神经网络算法相比具有避免局部最小,实现全局最优化,更好的泛化效果的优点,是高速公路事件检测的一种很有潜力的算法。     

基于动态分组的三维小波变换方法

为了实现高质量的视频压缩,文中提出了基于动态分组(DynamicGroupofFrame based)的三维小波变换方法(3D V Forward)。与以往的三维小波变换方法相比较,该方法能显著减少视频压缩解压缩过程的时间延迟,能避免基于固定帧分组的(GroupofFrame based)三维小波变换方法(3D GOF)对视频质量的不利影响,仿真结果表明了该方法的有效性。     

非高斯杂波协方差矩阵估计方法及CFAR特性分析

在独立同分布的球不变随机向量非高斯杂波背景下,针对采样协方差矩阵(SCM)、归一化采样协方差矩阵(NSCM)及相应的迭代估计矩阵(NSCM-RE),利用统一理论框架分析了3种不同估计矩阵对应的自适应归一化匹配滤波器(ANMF)的恒虚警率(CFAR)特性.理论分析表明,SCM-ANMF只对杂波归一化协方差矩阵具有CFAR特性;而NSCM-ANMF只对杂波功率水平具有CFAR特性;在有限次迭代下,NSCM-RE-ANMF对杂波归一化协方差矩阵不具有CFAR特性.为了改善ANMF的自适应特性,提出了一种自适应协方差矩阵估计方法(AE),并将其作为初始化矩阵进行迭代估计,在有限次迭代下,所获得的AE-RE-ANMF对杂波归一化协方差矩阵和功率水平均具有CFAR特性.最后,通过仿真验证了所提方法的有效性.     

Rational bases for convex polyhedra

In Wachspress (1975) [2] rational bases were constructed for convex polyhedra whose vertices were all of order three. The restriction to order three was first removed by Warren (1996) [3] and his analysis was refined subsequently by Warren and Schaefer (2004) [4]. A new algorithm (GADJ) for finding the denominator polynomial common to all the basis functions was exposed in Dasgupta and Wachspress (2007) [1] for convex polyhedra with all vertices of order three. This algorithm is applied here for generating bases for general convex polyhedra.     

Creep in Thin Magnetic Films†

In this paper a method of testing and realization of two-realizable threshold functions have been suggested. It is shown that the two–realizable functions may be factored into three different forms, (i) sum of or (ii) product of or (iii) ‘ sum–of–product ’ of threshold functions. Since threshold functions are unate functions, a method of decomposition of Boolean functions into unate functions in the above three forms is suggested. The concept of minimal unate function and the cover table of the given Boolean function und its complement is utilized to obtain the above three factored forms.     

Organizational synthesis in transboundary crises: Three principles for managing centralization and coordination in the corona virus crisis response

Boin (2019) argues that in transboundary crisis management it is almost impossible to achieve centralization and coordination. This article identifies three principles through which actors in a transboundary crisis can balance centralization with autonomy while shaping coordination along the way. We reanalysed three transboundary cases: the Dutch military mission in Afghanistan, the downing of MH17 and hurricane Irma striking Sint‐Maarten. The principles we found are as follows: (a) reformulating key strategic priorities, (b) flexible adaptation of crisis management protocols and (c) the emergence of multifunctional units. With these three principles, we reflect on challenges in the Dutch crisis response to the corona outbreak and propose improvements for progressing current crisis management efforts.     

Predicting proteinase specificities from free energy calculations

The role of the primary binding residue (P1) in complexes between three different subtilases (subtilisin Carlsberg, thermitase and proteinase K) and their canonical protein inhibitor eglin c have been studied by free energy calculations. Based on the crystal structures of eglin c in complex with subtilisin Carlsberg and thermitase, and a homology model of the eglin c-proteinase K complex, a total of 57 mutants have been constructed and docked into their host proteins. The binding free energy was then calculated using molecular dynamics (MD) simulations combined with the linear interaction energy (LIE) method for all complexes differing only in the nature of the amino acid at the P1 position. LIE calculations for 19 different complexes for each subtilase were thus carried out excluding proline. The effects of substitutions at the P1 position on the binding free energies are found to be very large, and positively charged residues (Arg, Lys and His) are particularly deleterious for all three enzymes. The charged variants of the acidic side chains are found to bind more favorably as compared to their protonated states in all three subtilases. Furthermore, hydrophobic amino acids are accommodated most favorably at the S1-site in all three enzymes. Comparison of the three series of binding free energies shows only minor differences in the 19 computed relative binding free energies among these subtilases. This is further reflected in the correlation coefficient between the 23 relative binding free energies obtained, including the possible protonation states of ionizable side chains, but excluding the P1 Pro, for subtilisin Carlsberg versus thermitase (0.95), subtilisin versus proteinase K (0.94) and thermitase versus proteinase K (0.96).     

系统决策的价值分析、价值寻求和价值综合集成的比较研究

基于元决策价值概念模型．对简单小系统、大系统、巨系统和复杂系统决策的价值问题进行研究。讨论了价值分析（VA）、价值寻求分析（VSA）、价值综合集成（VMS）3个层次的内涵;进而讨论了不同决策系统中决策与价值分析、价值寻求分析的关系．在此基础上基于复杂系统方法论,提出面向复杂系统问题决策的价值综合集成的思维框架。     

μ-Processor based capacitor testing and sorting

The three vital parameters of Plastic Film Capacitors are the Insulation Resistance (IR) Capacitance (C) and Dissipation Factor (DF-Tan δ). The paper presents a microprocessor based system for measuring the above three parameters and for sorting the capacitors depending upon their capacitance tolerance with respect to the standard.