Adaptive generalized predictive control algorithm implemented over an heterogeneous parallel architecture

In this paper a parallel implementation of an Adaprtive Generalized Predictive Control (AGPC) algorithm is presented. Since the AGPC algorithm needs to be fed with knowledge of the plant transfer function, the parallelization of a standard Recursive Least Squares (RLS) estimator and a GPC predictor is discussed here.

Distributed Diverging Topic Models: A Novel Algorithm for Large Scale Topic Modeling in Spark

Thesis (Master's)--University of Washington, 2014

The Salmon Algorithm - A New Population Based Search Metaheuristic

This thesis introduces the Salmon Algorithm, a search meta-heuristic which can be used for a variety of combinatorial optimization problems. This algorithm is loosely based on the path finding behaviour of salmon swimming upstream to spawn. There are a number of tunable parameters in the algorithm, so experiments were conducted to find the optimum parameter settings for different search spaces. The algorithm was tested on one instance of the Traveling Salesman Problem and found to have superior performance to an Ant Colony Algorithm and a Genetic Algorithm. It was then tested on three coding theory problems - optimal edit codes, optimal Hamming distance codes, and optimal covering codes. The algorithm produced improvements on the best known values for five of six of the test cases using edit codes. It matched the best known results on four out of seven of the Hamming codes as well as three out of three of the covering codes. The results suggest the Salmon Algorithm is competitive with established guided random search techniques, and may be superior in some search spaces.

A Multi-Objective Genetic Algorithm with Side Effect Machines for Motif Discovery

Understanding the machinery of gene regulation to control gene expression has been one of the main focuses of bioinformaticians for years. We use a multi-objective genetic algorithm to evolve a specialized version of side effect machines for degenerate motif discovery. We compare some suggested objectives for the motifs they find, test different multi-objective scoring schemes and probabilistic models for the background sequence models and report our results on a synthetic dataset and some biological benchmarking suites. We conclude with a comparison of our algorithm with some widely used motif discovery algorithms in the literature and suggest future directions for research in this area.

New Contig Creation Algorithm for the de novo DNA Assembly Problem

DNA assembly is among the most fundamental and difficult problems in bioinformatics. Near optimal assembly solutions are available for bacterial and small genomes, however assembling large and complex genomes especially the human genome using Next-Generation-Sequencing (NGS) technologies is shown to be very difficult because of the highly repetitive and complex nature of the human genome, short read lengths, uneven data coverage and tools that are not specifically built for human genomes. Moreover, many algorithms are not even scalable to human genome datasets containing hundreds of millions of short reads. The DNA assembly problem is usually divided into several subproblems including DNA data error detection and correction, contig creation, scaffolding and contigs orientation; each can be seen as a distinct research area. This thesis specifically focuses on creating contigs from the short reads and combining them with outputs from other tools in order to obtain better results. Three different assemblers including SOAPdenovo [Li09], Velvet [ZB08] and Meraculous [CHS+11] are selected for comparative purposes in this thesis. Obtained results show that this thesis’ work produces comparable results to other assemblers and combining our contigs to outputs from other tools, produces the best results outperforming all other investigated assemblers.

A Study of Ordered Gene Problems Featuring DNA Error Correction and DNA Fragment Assembly with a Variety of Heuristics, Genetic Algorithm Variations, and Dynamic Representations

Ordered gene problems are a very common classification of optimization problems. Because of their popularity countless algorithms have been developed in an attempt to find high quality solutions to the problems. It is also common to see many different types of problems reduced to ordered gene style problems as there are many popular heuristics and metaheuristics for them due to their popularity. Multiple ordered gene problems are studied, namely, the travelling salesman problem, bin packing problem, and graph colouring problem. In addition, two bioinformatics problems not traditionally seen as ordered gene problems are studied: DNA error correction and DNA fragment assembly. These problems are studied with multiple variations and combinations of heuristics and metaheuristics with two distinct types or representations. The majority of the algorithms are built around the Recentering- Restarting Genetic Algorithm. The algorithm variations were successful on all problems studied, and particularly for the two bioinformatics problems. For DNA Error Correction multiple cases were found with 100% of the codes being corrected. The algorithm variations were also able to beat all other state-of-the-art DNA Fragment Assemblers on 13 out of 16 benchmark problem instances.

Disease-Gene Association Using a Genetic Algorithm

Understanding the relationship between genetic diseases and the genes associated with them is an important problem regarding human health. The vast amount of data created from a large number of high-throughput experiments performed in the last few years has resulted in an unprecedented growth in computational methods to tackle the disease gene association problem. Nowadays, it is clear that a genetic disease is not a consequence of a defect in a single gene. Instead, the disease phenotype is a reflection of various genetic components interacting in a complex network. In fact, genetic diseases, like any other phenotype, occur as a result of various genes working in sync with each other in a single or several biological module(s). Using a genetic algorithm, our method tries to evolve communities containing the set of potential disease genes likely to be involved in a given genetic disease. Having a set of known disease genes, we first obtain a protein-protein interaction (PPI) network containing all the known disease genes. All the other genes inside the procured PPI network are then considered as candidate disease genes as they lie in the vicinity of the known disease genes in the network. Our method attempts to find communities of potential disease genes strongly working with one another and with the set of known disease genes. As a proof of concept, we tested our approach on 16 breast cancer genes and 15 Parkinson's Disease genes. We obtained comparable or better results than CIPHER, ENDEAVOUR and GPEC, three of the most reliable and frequently used disease-gene ranking frameworks.

Deep learning of representations and its application to computer vision

L’objectif de cette thèse par articles est de présenter modestement quelques étapes du parcours qui mènera (on espère) à une solution générale du problème de l’intelligence artificielle. Cette thèse contient quatre articles qui présentent chacun une différente nouvelle méthode d’inférence perceptive en utilisant l’apprentissage machine et, plus particulièrement, les réseaux neuronaux profonds. Chacun de ces documents met en évidence l’utilité de sa méthode proposée dans le cadre d’une tâche de vision par ordinateur. Ces méthodes sont applicables dans un contexte plus général, et dans certains cas elles on tété appliquées ailleurs, mais ceci ne sera pas abordé dans le contexte de cette de thèse. Dans le premier article, nous présentons deux nouveaux algorithmes d’inférence variationelle pour le modèle génératif d’images appelé codage parcimonieux “spike- and-slab” (CPSS). Ces méthodes d’inférence plus rapides nous permettent d’utiliser des modèles CPSS de tailles beaucoup plus grandes qu’auparavant. Nous démontrons qu’elles sont meilleures pour extraire des détecteur de caractéristiques quand très peu d’exemples étiquetés sont disponibles pour l’entraînement. Partant d’un modèle CPSS, nous construisons ensuite une architecture profonde, la machine de Boltzmann profonde partiellement dirigée (MBP-PD). Ce modèle a été conçu de manière à simplifier d’entraînement des machines de Boltzmann profondes qui nécessitent normalement une phase de pré-entraînement glouton pour chaque couche. Ce problème est réglé dans une certaine mesure, mais le coût d’inférence dans le nouveau modèle est relativement trop élevé pour permettre de l’utiliser de manière pratique. Dans le deuxième article, nous revenons au problème d’entraînement joint de machines de Boltzmann profondes. Cette fois, au lieu de changer de famille de modèles, nous introduisons un nouveau critère d’entraînement qui donne naissance aux machines de Boltzmann profondes à multiples prédictions (MBP-MP). Les MBP-MP sont entraînables en une seule étape et ont un meilleur taux de succès en classification que les MBP classiques. Elles s’entraînent aussi avec des méthodes variationelles standard au lieu de nécessiter un classificateur discriminant pour obtenir un bon taux de succès en classification. Par contre, un des inconvénients de tels modèles est leur incapacité de générer deséchantillons, mais ceci n’est pas trop grave puisque la performance de classification des machines de Boltzmann profondes n’est plus une priorité étant donné les dernières avancées en apprentissage supervisé. Malgré cela, les MBP-MP demeurent intéressantes parce qu’elles sont capable d’accomplir certaines tâches que des modèles purement supervisés ne peuvent pas faire, telles que celle de classifier des données incomplètes ou encore celle de combler intelligemment l’information manquante dans ces données incomplètes. Le travail présenté dans cette thèse s’est déroulé au milieu d’une période de transformations importantes du domaine de l’apprentissage à réseaux neuronaux profonds qui a été déclenchée par la découverte de l’algorithme de “dropout” par Geoffrey Hinton. Dropout rend possible un entraînement purement supervisé d’architectures de propagation unidirectionnel sans être exposé au danger de sur- entraînement. Le troisième article présenté dans cette thèse introduit une nouvelle fonction d’activation spécialement con ̧cue pour aller avec l’algorithme de Dropout. Cette fonction d’activation, appelée maxout, permet l’utilisation de aggrégation multi-canal dans un contexte d’apprentissage purement supervisé. Nous démontrons comment plusieurs tâches de reconnaissance d’objets sont mieux accomplies par l’utilisation de maxout. Pour terminer, sont présentons un vrai cas d’utilisation dans l’industrie pour la transcription d’adresses de maisons à plusieurs chiffres. En combinant maxout avec une nouvelle sorte de couche de sortie pour des réseaux neuronaux de convolution, nous démontrons qu’il est possible d’atteindre un taux de succès comparable à celui des humains sur un ensemble de données coriace constitué de photos prises par les voitures de Google. Ce système a été déployé avec succès chez Google pour lire environ cent million d’adresses de maisons.

A computer-algebraic approach to the study of the symmetry properties of molecules and clusters

This thesis work is dedicated to use the computer-algebraic approach for dealing with the group symmetries and studying the symmetry properties of molecules and clusters. The Maple package Bethe, created to extract and manipulate the group-theoretical data and to simplify some of the symmetry applications, is introduced. First of all the advantages of using Bethe to generate the group theoretical data are demonstrated. In the current version, the data of 72 frequently applied point groups can be used, together with the data for all of the corresponding double groups. The emphasize of this work is placed to the applications of this package in physics of molecules and clusters. Apart from the analysis of the spectral activity of molecules with point-group symmetry, it is demonstrated how Bethe can be used to understand the field splitting in crystals or to construct the corresponding wave functions. Several examples are worked out to display (some of) the present features of the Bethe program. While we cannot show all the details explicitly, these examples certainly demonstrate the great potential in applying computer algebraic techniques to study the symmetry properties of molecules and clusters. A special attention is placed in this thesis work on the flexibility of the Bethe package, which makes it possible to implement another applications of symmetry. This implementation is very reasonable, because some of the most complicated steps of the possible future applications are already realized within the Bethe. For instance, the vibrational coordinates in terms of the internal displacement vectors for the Wilson's method and the same coordinates in terms of cartesian displacement vectors as well as the Clebsch-Gordan coefficients for the Jahn-Teller problem are generated in the present version of the program. For the Jahn-Teller problem, moreover, use of the computer-algebraic tool seems to be even inevitable, because this problem demands an analytical access to the adiabatic potential and, therefore, can not be realized by the numerical algorithm. However, the ability of the Bethe package is not exhausted by applications, mentioned in this thesis work. There are various directions in which the Bethe program could be developed in the future. Apart from (i) studying of the magnetic properties of materials and (ii) optical transitions, interest can be pointed out for (iii) the vibronic spectroscopy, and many others. Implementation of these applications into the package can make Bethe a much more powerful tool.

A concurrent region growing algorithm guided by circumscribed contours

Image segmentation of natural scenes constitutes a major problem in machine vision. This paper presents a new proposal for the image segmentation problem which has been based on the integration of edge and region information. This approach begins by detecting the main contours of the scene which are later used to guide a concurrent set of growing processes. A previous analysis of the seed pixels permits adjustment of the homogeneity criterion to the region's characteristics during the growing process. Since the high variability of regions representing outdoor scenes makes the classical homogeneity criteria useless, a new homogeneity criterion based on clustering analysis and convex hull construction is proposed. Experimental results have proven the reliability of the proposed approach

Proposal of a parallel architecture for a motion detection algorithm

This paper proposes a parallel architecture for estimation of the motion of an underwater robot. It is well known that image processing requires a huge amount of computation, mainly at low-level processing where the algorithms are dealing with a great number of data. In a motion estimation algorithm, correspondences between two images have to be solved at the low level. In the underwater imaging, normalised correlation can be a solution in the presence of non-uniform illumination. Due to its regular processing scheme, parallel implementation of the correspondence problem can be an adequate approach to reduce the computation time. Taking into consideration the complexity of the normalised correlation criteria, a new approach using parallel organisation of every processor from the architecture is proposed

Speculative parallelization of multipath radiosity algorithm

In computer graphics, global illumination algorithms take into account not only the light that comes directly from the sources, but also the light interreflections. This kind of algorithms produce very realistic images, but at a high computational cost, especially when dealing with complex environments. Parallel computation has been successfully applied to such algorithms in order to make it possible to compute highly-realistic images in a reasonable time. We introduce here a speculation-based parallel solution for a global illumination algorithm in the context of radiosity, in which we have taken advantage of the hierarchical nature of such an algorithm

A Monte Carlo-Based Fiber Tracking Algorithm using Diffusion Tensor MRI

Diffusion tensor magnetic resonance imaging, which measures directional information of water diffusion in the brain, has emerged as a powerful tool for human brain studies. In this paper, we introduce a new Monte Carlo-based fiber tracking approach to estimate brain connectivity. One of the main characteristics of this approach is that all parameters of the algorithm are automatically determined at each point using the entropy of the eigenvalues of the diffusion tensor. Experimental results show the good performance of the proposed approach

A geometric algorithm for an artificial life form based on anachronistic methods

Modern methods of spawning new technological motifs are not appropriate when it is desired to realize artificial life as an actual real world entity unto itself (Pattee 1995; Brooks 2006; Chalmers 1995). Many fundamental aspects of such a machine are absent in common methods, which generally lack methodologies of construction. In this paper we mix classical and modern studies in order to attempt to realize an artificial life form from first principles. A model of an algorithm is introduced, its methodology of construction is presented, and the fundamental source from which it sprang is discussed.

CAMCR: Computer assisted mixture model analysis for capture-recapture count data

Population size estimation with discrete or nonparametric mixture models is considered, and reliable ways of construction of the nonparametric mixture model estimator are reviewed and set into perspective. Construction of the maximum likelihood estimator of the mixing distribution is done for any number of components up to the global nonparametric maximum likelihood bound using the EM algorithm. In addition, the estimators of Chao and Zelterman are considered with some generalisations of Zelterman’s estimator. All computations are done with CAMCR, a special software developed for population size estimation with mixture models. Several examples and data sets are discussed and the estimators illustrated. Problems using the mixture model-based estimators are highlighted.