Computer Sciences Commons^™

Nonlinear Constrained Optimizer And Parallel Processing For Golden Block Line Search, Duc T. Nguyen, Wilson H. Tang, Yeou K. Tung, Hakizumwami B. Runesha

Civil & Environmental Engineering Faculty Publications

Generalized exponential penalty functions are constructed for the multiplier methods in solving nonlinear programming problems. The non-smooth extreme constraint Gext is replaced by a single smooth constraint Gs by using the generalized exponential function (base a>1). The well-known K.S. function is found to be a special case of our proposed formulation. Parallel processing for Golden block line search algorithm is then summarized, which can also be integrated into our formulation. Both small and large-scale nonlinear programming problems (up to 2000 variables and 2000 nonlinear constraints) have been solved to validate the proposed algorithms.

Automatic Target Cueing Of Hyperspectral Image Data, Terry A. Wilson

Theses and Dissertations

Modern imaging sensors produce vast amounts data, overwhelming human analysts. One such sensor is the Airborne Visible and Infrared Imaging Spectrometer (AVIRIS) hyperspectral sensor. The AVIRIS sensor simultaneously collects data in 224 spectral bands that range from 0.4µm to 2.5µm in approximately 10nm increments, producing 224 images, each representing a single spectral band. Autonomous systems are required that can fuse "important" spectral bands and then classify regions of interest if all of this data is to be exploited. This dissertation presents a comprehensive solution that consists of a new physiologically motivated fusion algorithm and a novel Bayes optimal self-architecting classifier …

Representations, Approximations, And Algorithms For Mathematical Speech Processing, Laura R. Suzuki

Theses and Dissertations

Representing speech signals such that specific characteristics of speech are included is essential in many Air Force and DoD signal processing applications. A mathematical construct called a frame is presented which captures the important time-varying characteristic of speech. Roughly speaking, frames generalize the idea of an orthogonal basis in a Hilbert space, Specific spaces applicable to speech are L₂(R) and the Hardy spaces H_p(D) for p> 1 where D is the unit disk in the complex plane. Results are given for representations in the Hardy spaces involving Carleson's inequalities (and its extensions), …

New Algorithms For Moving-Bank Multiple Model Adaptive Estimation, Juan R. Vasquez

Theses and Dissertations

The focus of this research is to provide methods for generating precise parameter estimates in the face of potentially significant parameter variations such as system component failures. The standard Multiple Model Adaptive Estimation (MMAE) algorithm uses a bank of Kalman filters, each based on a different model of the system. A new moving-bank MMAE algorithm is developed based on exploitation of the density data available from the MMAE. The methods used to exploit this information include various measures of the density data and a decision-making logic used to move, expand, and contract the MMAE bank of filters. Parameter discretization within …

Optimal Contention-Free Unicast-Based Multicasting In Switch-Based Networks Of Workstations, Ran Libeskind-Hadas, Dominic Mazzoni '99, Ranjith Rajagopalan '99

All HMC Faculty Publications and Research

A unicast-based multicasting algorithm is presented for arbitrary interconnection networks arising in switch-based networks of workstations. The algorithm is optimal with respect tot he number of startups incurred and is provably free from depth contention. Specifically, no two constituent unicasts for the same multicast contend for a common channel, even if some unicasts are delayed due to unpredictable variations in latencies. The algorithm uses an underlying partially adaptive deadlock-free unicast routing algorithm. Simulation results indicate that the algorithm behaves as predicted by its theoretical properties and provides a promising approach to unicast-based multicasting.

Sorting In Parallel, Ran Libeskind-Hadas

All HMC Faculty Publications and Research

In 1842, L.F. Menabrea anticipated the benefits of parallel computing in an article that appeared in the Swiss Journal Bibliotheque universelle de Geneve:

When a long series of identical computations is to be performed, such as those required for the formation of numerical tables, the machine can be brought into play so as to give several results at the same time, which will greatly abridge the whole amount of the processes.

Although more than a century passed before Menabrea's vision became a reality, today parallel computers with hundreds and even thousands of processors are used in a broad range …

Tree-Based Multicasting In Wormhole-Routed Irregular Topologies, Ran Libeskind-Hadas, Dominic Mazzoni '99, Ranjith Rajagopalan '99

All HMC Faculty Publications and Research

A deadlock-free tree-based multicast routing algorithm is presented for all direct networks, regardless of interconnection topology. The algorithm delivers a message to any number of destinations using only a single startup phase. In contrast to existing tree-based schemes, this algorithm applies to all interconnection topologies, requires only fixed-sized input buffers that are independent of maximum message length, and uses a single asynchronous flit replication mechanism. The theoretical basis of the technique used here is sufficiently general to develop other tree-based multicasting algorithms for regular and irregular topologies. Simulation results demonstrate that this tree-based algorithm provides a very promising means of …

Concept Vectors: A Synthesis Of Concept Mapping And Matrices For Knowledge Representation In Intelligent Tutoring Systems, Mark L. Dyson

Theses and Dissertations

A review of the literature relating to intelligent tutoring systems (ITS) reveals that the bulk of research to date is focused on the student, and on methods for representing the knowledge itself. From student models to learning schemas to presentation methods, comparatively little attention has been paid to the problem of educators attempting to build viable lesson plans for use in an ITS environment--yet when this problem is addressed in the literature, it is recognized as a potentially daunting one. This thesis addresses the problem of ITS lesson plan development by proposing a practical, computable approach for knowledge engineering that …

A Single Chip Low Power Implementation Of An Asynchronous Fft Algorithm For Space Applications, Bruce W. Hunt

Theses and Dissertations

A fully asynchronous fixed point FFT processor is introduced for low power space applications. The architecture is based on an algorithm developed by Suter and Stevens specifically for a low power implementation. The novelty of this architecture lies in its high localization of components and pipelining with no need to share a global memory. High throughput is attained using large numbers of small, local components working in parallel. A derivation of the algorithm from the discrete Fourier transform is presented followed by a discussion of circuit design parameters specifically, those relevant to space applications. The generic architecture is explained with …

Modeling And Simulation Support For Parallel Algorithms In A High-Speed Network, Dustin E. Yates

Theses and Dissertations

This thesis investigates the ability of a simulation model to compare and contrast parallel processing algorithms in a high-speed network. The model extends existing modeling, analysis, and comparison of parallel algorithms by providing graphics based components that facilitate the measurement of system resources. Simulation components are based on the Myrinet local area network standard. The models provide seven different topologies to contrast the performance of five variations of Fast Fourier Transform (FFT) algorithms. Furthermore, the models were implemented using a commercially developed product that facilitates the testing of additional topologies and the investigation of hardware variations. Accurate comparisons are statistically …

Applications Of Unsupervised Clustering Algorithms To Aircraft Identification Using High Range Resolution Radar, Dzung Tri Pham

Theses and Dissertations

Identification of aircraft from high range resolution (HRR) radar range profiles requires a database of information capturing the variability of the individual range profiles as a function of viewing aspect. This database can be a collection of individual signatures or a collection of average signatures distributed over the region of viewing aspect of interest. An efficient database is one which captures the intrinsic variability of the HRR signatures without either excessive redundancy typical of single-signature databases, or without the loss of information common when averaging arbitrary groups of signatures. The identification of 'natural' clustering of similar HRR signatures provides a …

Adaptive Multicast Routing In Wormhole Networks, Ran Libeskind-Hadas, Tom Hehre '96, Andrew Hutchings '98, Mark Reyes '98, Kevin Watkins '97

All HMC Faculty Publications and Research

Multicast communication has applications in a number of fundamental operations in parallel computing. An effective multicast routing algorithm must be free from both livelock and deadlock while minimizing communication latency. We describe two classes of multicast wormhole routing algorithms that employ the multi-destination wormhole hardware mechanism proposed by Lin et al. [12] and Panda et al. [17]. Specific examples of these classes of algorithms are described and experimental results suggests that such algorithms enjoy low communication latencies across a range of network loads.

Random Number Generators For Parallel Computers, Paul D. Coddington

Northeast Parallel Architecture Center

Random number generators are used in many applications, from slot machines to simulations of nuclear reactors. For many computational science applications, such as Monte Carlo simulation, it is crucial that the generators have good randomness properties. This is particularly true for large-scale simulations done on high-performance parallel computers. Good random number generators are hard to find, and many widely-used techniques have been shown to be inadequate. Finding high-quality, efficient algorithms for random number generation on parallel computers is even more difficult. Here we present a review of the most commonly-used random number generators for parallel computers, and evaluate each generator …

A Compiler Algorithm For Optimizing Locality In Loop Nests, Mahmut Kandemir, J. Ramanujam, Alok Choudhary

Electrical Engineering and Computer Science - All Scholarship

This paper describes an algorithm to optimize cache locality in scientific codes on uniprocessor and multiprocessor machines. A distinctive characteristic of our algorithm is that it considers loop and data layout transformations in a unified framework. We illustrate through examples that our approach is very effective at reducing cache misses and tile size sensitivity of blocked loop nests; and can optimize nests for which optimization techniques based on loop transformations alone are not successful. An important special case is the one in which data layouts of some arrays are fixed and cannot be changed. We show how our algorithm can …

Single-Layer Channel Routing And Placement With Single-Sided Nets, Ronald I. Greenberg, Jau-Der Shih

Computer Science: Faculty Publications and Other Works

This paper considers the optimal offset, feasible offset, and optimal placement problems for a more general form of single-layer VLSI channel routing than has usually been considered in the past. Most prior works require that every net has exactly one terminal on each side of the channel. As long as only one side of the channel contains multiple terminals of the same net, we provide linear-time solutions to all three problems. Such results are implausible if the placement of terminals is entirely unrestricted; in fact, the size of the output for the feasible offset problem may be Ω(n^2). The linear-time …

Variable Step Size Lms Adaptive Filters With Delayed Coefficient Updating, Guixian Xu

Electrical & Computer Engineering Theses & Dissertations

A new approach to delayed LMS adaptive filtering is presented, which uses a variable step size for coefficient updating to increase convergence speed and improve tracking characteristics. The new algorithm, called delayed variable step size LMS (DVLMS), is explained, analyzed, and simulated to experimentally determine performance characteristics. Three different strategies for adjusting the step size are examined, and their performance is compared. Also, simulation results are presented to show that the proposed DVLMS systems provide faster convergence and lower mis-adjustment than previously proposed DLMS systems.

Low-Degree Spanning Trees Of Small Weight, Samir Khuller, Balaji Raghavachari, Neal Young

Dartmouth Scholarship

Given n points in the plane, the degree-K spanning-tree problem asks for a spanning tree of minimum weight in which the degree of each vertex is at most K. This paper addresses the problem of computing low-weight degree-K spanning trees for $K > 2$. It is shown that for an arbitrary collection of n points in the plane, there exists a spanning tree of degree 3 whose weight is at most 1.5 times the weight of a minimum spanning tree. It is shown that there exists a spanning tree of degree 4 whose weight is at most 1.25 times …

Unsupervised Algorithms For Learning Emergent Spatio-Temporal Correlations, Chaitanya Tumuluri

Electrical Engineering and Computer Science - Technical Reports

Many applications require the extraction of spatiotemporal correlations among dynamically emergent features of non-stationary distributions. In such applications it is not possible to obtain an a priori analytical characterization of the emergent distribution. This paper extends the Growing Cell Structures (GCS) network and presents two novel (GIST and GEST) networks, which combine unsupervised feature-extraction and Hebbian learning, for tracking such emergent correlations. The networks were successfully tested on the challenging Data Mapping problem, using an execution driven simulation of their implementation in hardware. The results of the simulations show the successful use of the GIST and GEST networks for extracting …

Torus Routing In The Presence Of Multicasts, Hiroki Ishibashi

Theses Digitization Project

No abstract provided.

Open-Loop State-Space Model Identification From Closed-Loop Data, Lori Guy

Mechanical & Aerospace Engineering Theses & Dissertations

This thesis provides an investigation of a system identification algorithm which identifies an open-loop state-space model from a linear system that is operating under closed-loop conditions. In order to investigate the system identification algorithm some basic ideas of system identification theory are reviewed. Examples using simulated data are presented to characterize the effects of varying the parameters for open-loop and closed-loop system identification processes. Both noise· free and noise contaminated cases are simulated. Linear Quadratic Gaussian (LQG) control theory is reviewed and the motivation for using iterative LQG control feedback is discussed. The derivation of the proposed system identification algorithm …

Approximation Algorithms: Good Solutions To Hard Problems, Ran Libeskind-Hadas

All HMC Faculty Publications and Research

Consider a computer network represented by an undirected graph where the vertices represent computer nodes and the edges represent links between the nodes. Since some of the links in the network may become faulty, link testing devices are placed at some of the nodes. A tester at a particular node can test all links incident to that node. Since the testers are expensive, however, we wish to deploy the minimum number of these devices such that every link is incidient to at least one node containing a tester. In graph theoretic terms, a vertex cover is a subset of the …

Automatic Pcb Inspection Systems, M. Moganti, Fikret Erçal

Computer Science Faculty Research & Creative Works

There are more than 50 process steps required to fabricate a printed circuit board (PCB). To ensure quality, human operators simply inspect the work visually against prescribed standards. The decisions made by this labor intensive, and therefore costly, procedure often also involve subjective judgements. Automatic inspection systems remove the subjective aspects and provide fast, quantitative dimensional assessments. Machine vision may answer the manufacturing industry's need to improve product quality and increase productivity. The major limitation of existing inspection systems is that all the algorithms need a special hardware platform to achieve the desired real-time speeds. This makes the systems extremely …

A Recursive Least Squares Training Algorithm For Multilayer Recurrent Neural Networks, Q. Xu, K. Krishnamurthy, Bruce M. Mcmillin, Wen Feng Lu

Mechanical and Aerospace Engineering Faculty Research & Creative Works

Recurrent neural networks have the potential to perform significantly better than the commonly used feedforward neural networks due to their dynamical nature. However, they have received less attention because training algorithms/architectures have not been well developed. In this study, a recursive least squares algorithm to train recurrent neural networks with an arbitrary number of hidden layers is developed. The training algorithm is developed as an extension of the standard recursive estimation problem. Simulated results obtained for identification of the dynamics of a nonlinear dynamical system show promising results.

Identification Of Cutting Force In End Milling Operations Using Recurrent Neural Networks, Q. Xu, K. Krishnamurthy, Bruce M. Mcmillin, Wen Feng Lu

Mechanical and Aerospace Engineering Faculty Research & Creative Works

The problem of identifying the cutting force in end milling operations is considered in this study. Recurrent neural networks are used here and are trained using a recursive least squares training algorithm. Training results for data obtained from a SAJO 3-axis vertical milling machine for steady slot cuts are presented. The results show that a recurrent neural network can learn the functional relationship between the feed rate and steady-state average resultant cutting force very well. Furthermore, results for the Mackey-Glass time series prediction problem are presented to illustrate the faster learning capability of the neural network scheme presented here

Knowledge-Based Nonuniform Crossover, Harpal Maini, Kishan Mehrotra, Chilukuri K. Mohan, Sanjay Ranka

Electrical Engineering and Computer Science - Technical Reports

We present a new "knowledge-based non-uniform crossover" (KNUX) operator for genetic algorithms (GA's) that generalizes uniform crossover. We extend this to "Dynamic KNUX" (DKNUX), which constantly updates the knowledge extracted so far from the environment's feedback on previously generated chromosomes. KNUX can improve on good solutions previously obtained by using other algorithms. The modifications made by KNUX are orthogonal to other changes in parameters of GA's, and can be pursued together with any other proposed improvements. Whereas most genetic search methods focus on improving the move-selection procedures, after having chosen a fixed move-generation mechanism, KNUX and DKNUX make the move-generation …

The Transportation Primitive, Ravi V. Shankar, Khaled A. Alsabti, Sanjay Ranka

College of Engineering and Computer Science - Former Departments, Centers, Institutes and Projects

This paper presents algorithms for implementing the transportation primitive on a distributed memory parallel architecture. The transportation primitive performs many-to-many personalized communication with bounded incoming and outgoing traffic. We present a two-stage deterministic algorithm that decomposes the communication with possibly high variance in message size into two communication stages with low message size variance. If the maximum outgoing or incoming traffic at any processor is t, transportation can be done in 2t¯ time (+ lower order terms) when t O(p 2 + pø=¯) (¯ is the inverse of the data transfer rate, ø is the startup overhead). If the maximum …

Runtime Array Redistribution In Hpf Programs, Rajeev Thakur, Alok Choudhary, Geoffrey C. Fox

Northeast Parallel Architecture Center

This paper describes efficient algorithms for runtime array redistribution in HPF programs. We consider block(m) to cyclic, cyclic to block(m) and the general cyclic(x) to cyclic(y) type redistributions. We initially describe algorithms for one-dimensional arrays and then extend the methodology to multidimensional arrays. The algorithms are practical enough to be easily implemented in the runtime library of an HPF compiler and can also be directly used in application programs requiring redistribution. Performance results on the Intel Paragon are discussed.

A Numerical Study Of High-Speed Missile Configurations Using A Block- Structured Parallel Algorithm, Douglas C. Blake

Theses and Dissertations

A numerical analysis of the aerodynamic phenomena associated with the high-speed flight of a sharp-nosed, four-finned, high-fineness ratio missile using a block-structured, parallel computer algorithm is presented. The algorithm, PANS-3EM, utilizes a second-order-accurate, shock-capturing, Total Variation Diminishing scheme and incorporates a Baldwin-Lomax turbulence model. PANS-3EM allows for extreme flexibility in the choice of computational domain decomposition and computing machine of implementation. Developmental work consists of conceptualization and verification of the algorithm as well as parallel performance and scalability studies conducted on a variety of computing platforms. Using PANS-3EM, the aerodynamic characteristics of the missile are investigated. Drag and pitching moment …

Genetic Algorithms For Stochastic Flow Shop No Wait Scheduling, Harpal Maini, Ubirajara R. Ferreira

Electrical Engineering and Computer Science - Technical Reports

ln this paper we present Genetic Algorithms - evolutionary algorithms based on an analogy with natural selection and survival of the fittest - applied to an NP Complete combinatorial optimization problem: minimizing the makespan of a Stochastic Flow Shop No Wait (FSNW) schedule. This is an important optimization criteria in real-world situations and the problem itself is of practical significance. We restrict our applications to the three machine flow shop no wait problem which is known to be NP complete. The stochastic hypothesis is that the processing times of jobs are described by normally distributed random variables. We discuss how …

Design And Implementation Of Two Text Recognition Algorithms, Madhumathi Yendamuri

Theses

This report presents two algorithms for text recognition. One is a neural-based orthogonal vector with pseudo-inverse approach for pattern recognition. A method to generate N orthogonal vectors for an N-neuron network is also presented. This approach converges the input to the corresponding orthogonal vector representing the prototype vector. This approach can restore an image to the original image and thus has error recovery capacility. Also, the concept of sub-networking is applied to this approach to enhance the memory capacity of the neural network. This concept drastically increases the memory capacity of the network and also causes a reduction of the …