Dissertations / Theses on the topic 'Classifier systems'

Two methods of designing a point classifier are discussed in this paper, one is a binary decision tree classifier based on the Fisher's linear discriminant function as a decision rule at each nonterminal node, and the other is a contextual classifier which gives each pixel the highest probability label given some substantially sized context including the pixel. Experiments were performed both on a simulated image and real images to illustrate the improvement of the classification accuracy over the conventional single-stage Bayes classifier under Gaussian distribution assumption.
Master of Science

The ability to learn from experience is a key aspect of intelligence. Incorporating this ability into a computer is a formidable problem. Genetic algorithms coupled to learning classifier systems are powerful tools for tackling this task. While genetic algorithms can be shown to be near optimal solutions for the search task they perform, no similar proof exists for classifier systems. My research investigated two aspects of classifier systems, classifier selection and credit assignment. Explicit world models, look ahead and incremental planning are incorporated into the classifier system framework in order to make use of more of the information available to the system, and a more sophisticated approach to credit assignment is attempted. The investigation involved the construction of four different classifier systems, and testing each of these systems in three separate virtual worlds. Wilson's Animat research was carefully reconstructed, and used as the control in a scientific experiment testing the efficacy of the various strategies embodied in three experimental systems. The three experimental classifier systems all contained explicit world models and lookahead. One was an extension of Wilson's Animat, the other two involved an entirely new credit assignment scheme inspired by Watkins's Q-learning technique. Use of this technique enabled the incorporation of an incremental planner, similar to Sutton's Dyna-Q research, into one of the classifier systems, distinguishing it from the other Q-learning based classifier system. The research shows that use of explicit world models and lookahead significantly decreases the time required in order to discoverr paths to well rewarded goals. It also shows that incremental planning can be used to further increase learning speed. While the experimental classifier systems were quick at discovery, they did not necessarily exploit these discoveries. Because of this, the performance of these systems was variable between virtual worlds. The experimental systems were outperformed by the control in two out of the three virtual worlds investigated. Nevertheless, the value of explicit world models and lookahead within classifier systems is established. Likewise, the adaptation of Q-learning and incremental planning to classifier systems has been successfully demonstrated.

Els grans avenços en el camp de l’aprenentatge automàtic han resultat en el disseny de màquines competents que són capaces d’aprendre i d’extreure informació útil i original de l’experiència. Recentment, algunes d’aquestes tècniques d’aprenentatge s’han aplicat amb èxit per resoldre problemes del món real en àmbits tecnològics, mèdics, científics i industrials, els quals no es podien tractar amb tècniques convencionals d’anàlisi ja sigui per la seva complexitat o pel gran volum de dades a processar. Donat aquest èxit inicial, actualment els sistemes d’aprenentatge s’enfronten a problemes de complexitat més elevada, el que ha resultat en un augment de l’activitat investigadora entorn sistemes capaços d’afrontar nous problemes del món real eficientment i de manera escalable. Una de les famílies d’algorismes més prometedores en l’aprenentatge automàtic són els sistemes classificadors basats en algorismes genetics (LCSs), el funcionament dels quals s’inspira en la natura. Els LCSs intenten representar les polítiques d’actuació d’experts humans amb un conjunt de regles que s’empren per escollir les millors accions a realitzar en tot moment. Així doncs, aquests sistemes aprenen polítiques d’actuació de manera incremental a mida que van adquirint experiència a través de la informació nova que se’ls va presentant durant el temps. Els LCSs s’han aplicat, amb èxit, a camps tan diversos com la predicció de càncer de pròstata o el suport a la inversió en borsa, entre altres. A més en alguns casos s’ha demostrat que els LCSs realitzen tasques superant la precisió dels éssers humans. El propòsit d’aquesta tesi és explorar la naturalesa de l’aprenentatge online dels LCSs d’estil Michigan per a la mineria de grans quantitats de dades en forma de fluxos d’informació continus a alta velocitat i canviants en el temps. Molt sovint, l’extracció de coneixement a partir d’aquestes fonts de dades és clau per tal d’obtenir una millor comprensió dels processos que les dades estan descrivint. Així, aprendre d’aquestes dades planteja nous reptes a les tècniques tradicionals d’aprenentatge automàtic, les quals no estan dissenyades per tractar fluxos de dades continus i on els conceptes i els nivells de soroll poden variar amb el temps de forma arbitrària. La contribució de la present tesi pren l’eXtended Classifier System (XCS), el LCS d’estil Michigan més estudiat i un dels algoritmes d’aprenentatge automàtic més competents, com el punt de partida. D’aquesta manera els reptes abordats en aquesta tesi són dos: el primer desafiament és la construcció d’un sistema supervisat competent sobre el framework dels LCSs d’estil Michigan que aprèn dels fluxos de dades amb una capacitat de reacció ràpida als canvis de concepte i entrades amb soroll. Com moltes aplicacions científiques i industrials generen grans quantitats de dades sense etiquetar, el segon repte és aplicar les lliçons apreses per continuar amb el disseny de LCSs d’estil Michigan capaços de solucionar problemes online sense assumir una estructura a priori en els dades d’entrada.
Los grandes avances en el campo del aprendizaje automático han resultado en el diseño de máquinas capaces de aprender y de extraer información útil y original de la experiencia. Recientemente alguna de estas técnicas de aprendizaje se han aplicado con éxito para resolver problemas del mundo real en ámbitos tecnológicos, médicos, científicos e industriales, los cuales no se podían tratar con técnicas convencionales de análisis ya sea por su complejidad o por el gran volumen de datos a procesar. Dado este éxito inicial, los sistemas de aprendizaje automático se enfrentan actualmente a problemas de complejidad cada vez m ́as elevada, lo que ha resultado en un aumento de la actividad investigadora en sistemas capaces de afrontar nuevos problemas del mundo real de manera eficiente y escalable. Una de las familias más prometedoras dentro del aprendizaje automático son los sistemas clasificadores basados en algoritmos genéticos (LCSs), el funcionamiento de los cuales se inspira en la naturaleza. Los LCSs intentan representar las políticas de actuación de expertos humanos usando conjuntos de reglas que se emplean para escoger las mejores acciones a realizar en todo momento. Así pues estos sistemas aprenden políticas de actuación de manera incremental mientras van adquiriendo experiencia a través de la nueva información que se les va presentando. Los LCSs se han aplicado con éxito en campos tan diversos como en la predicción de cáncer de próstata o en sistemas de soporte de bolsa, entre otros. Además en algunos casos se ha demostrado que los LCSs realizan tareas superando la precisión de expertos humanos. El propósito de la presente tesis es explorar la naturaleza online del aprendizaje empleado por los LCSs de estilo Michigan para la minería de grandes cantidades de datos en forma de flujos continuos de información a alta velocidad y cambiantes en el tiempo. La extracción del conocimiento a partir de estas fuentes de datos es clave para obtener una mejor comprensión de los procesos que se describen. Así, aprender de estos datos plantea nuevos retos a las técnicas tradicionales, las cuales no están diseñadas para tratar flujos de datos continuos y donde los conceptos y los niveles de ruido pueden variar en el tiempo de forma arbitraria. La contribución del la presente tesis toma el eXtended Classifier System (XCS), el LCS de tipo Michigan más estudiado y uno de los sistemas de aprendizaje automático más competentes, como punto de partida. De esta forma los retos abordados en esta tesis son dos: el primer desafío es la construcción de un sistema supervisado competente sobre el framework de los LCSs de estilo Michigan que aprende de flujos de datos con una capacidad de reacción rápida a los cambios de concepto y al ruido. Como muchas aplicaciones científicas e industriales generan grandes volúmenes de datos sin etiquetar, el segundo reto es aplicar las lecciones aprendidas para continuar con el diseño de nuevos LCSs de tipo Michigan capaces de solucionar problemas online sin asumir una estructura a priori en los datos de entrada.
Last advances in machine learning have fostered the design of competent algorithms that are able to learn and extract novel and useful information from data. Recently, some of these techniques have been successfully applied to solve real-­‐world problems in distinct technological, scientific and industrial areas; problems that were not possible to handle by the traditional engineering methodology of analysis either for their inherent complexity or by the huge volumes of data involved. Due to the initial success of these pioneers, current machine learning systems are facing problems with higher difficulties that hamper the learning process of such algorithms, promoting the interest of practitioners for designing systems that are able to scalably and efficiently tackle real-­‐world problems. One of the most appealing machine learning paradigms are Learning Classifier Systems (LCSs), and more specifically Michigan-­‐style LCSs, an open framework that combines an apportionment of credit mechanism with a knowledge discovery technique inspired by biological processes to evolve their internal knowledge. In this regard, LCSs mimic human experts by making use of rule lists to choose the best action to a given problem situation, acquiring their knowledge through the experience. LCSs have been applied with relative success to a wide set of real-­‐ world problems such as cancer prediction or business support systems, among many others. Furthermore, on some of these areas LCSs have demonstrated learning capacities that exceed those of human experts for that particular task. The purpose of this thesis is to explore the online learning nature of Michigan-­‐style LCSs for mining large amounts of data in the form of continuous, high speed and time-­‐changing streams of information. Most often, extracting knowledge from these data is key, in order to gain a better understanding of the processes that the data are describing. Learning from these data poses new challenges to traditional machine learning techniques, which are not typically designed to deal with data in which concepts and noise levels may vary over time. The contribution of this thesis takes the extended classifier system (XCS), the most studied Michigan-­‐style LCS and one of the most competent machine learning algorithms, as the starting point. Thus, the challenges addressed in this thesis are twofold: the first challenge is building a competent supervised system based on the guidance of Michigan-­‐style LCSs that learns from data streams with a fast reaction capacity to changes in concept and noisy inputs. As many scientific and industrial applications generate vast amounts of unlabelled data, the second challenge is to apply the lessons learned in the previous issue to continue with the design of unsupervised Michigan-­‐style LCSs that handle online problems without assuming any a priori structure in input data.

This thesis investigates the use of self-adaptation and neural constructivism within a neural Learning Classifier System framework. The system uses a classifier structure whereby each classifier condition is represented by an artificial neural network, which is used to compute an action in response to an environmental stimulus. We implement this neural representation in two modem Learning Classifier Systems, XCS and XCSF. A classic problem in neural networks revolves around network topology considerations; how many neurons should the network consist of? How should we configure their topological arrangement and inter-neural connectivity patterns to ensure high performance? Similarly in Learning Classifier Systems, hand-tuning of parameters is sometimes necessary to achieve acceptable system performance. We employ a number of mechanisms to address these potential deficiencies. Neural Constructivism is utilised to automatically alter network topology to reflect the complexity of the environment. It is shown that appropriate internal classifier complexity emerges during learning at a rate controlled by the learner. The resulting systems are applied to real-valued, noisy simulated maze environments and a simulated robotics platform. The main areas of novelty include the first use of self-adaptive constructivism within XCSF, the first implementation of temporally-sensitive spiking classifier representations within this constructive XC SF, and the demonstration of temporal functionality of such representations in noisy continuous-valued and robotic environments.

Learning Classifier Systems (LCS) traditionally use a ternary encoding to generalise over the environmental inputs and to associate appropriate actions. However, a number of schemes have been presented beyond this, ranging from integers to artificial neural networks. This thesis investigates the use of Dynamical Genetic Programming (DGP) as a knowledge representation within LCS. DGP is a temporally dynamic, graph-based, symbolic representation. Temporal dynamism has been identified as an important aspect in biological systems, artificial life, and cognition in general. Furthermore, discrete dynamical systems have been found to exhibit inherent content-addressable memory. In this thesis, the collective emergent behaviour of ensembles of such dynamical function networks are herein shown to be exploitable toward solving various computational tasks. Significantly, it is shown possible to exploit the variable-length, adaptive memory existing inherently within the networks under an asynchronous scheme, and where all new parameters introduced are self-adaptive. It is shown possible to exploit the collective mechanics to solve both discrete and continuous-valued reinforcement learning problems, and to perform symbolic regression. In particular, the representation is shown to provide improved performance beyond a traditional Genetic Programming benchmark on a number of a composite polynomial regression tasks. Superior performance to previously published techniques is also shown in a continuous-input-output reinforcement learning problem. Finally, it is shown possible to perform multi-step-ahead predictions of a financial time-series by repeatedly sampling the network states at succeeding temporal intervals.

Learning Classifier Systems (LCS) are a family of rule-based machine learning methods. They aim at the autonomous production of potentially human readable results that are the most compact generalised representation whilst also maintaining high predictive accuracy, with a wide range of application areas, such as autonomous robotics, economics, and multi-agent systems. Their design is mainly approached heuristically and, even though their performance is competitive in regression and classification tasks, they do not meet their expected performance in sequential decision tasks despite being initially designed for such tasks. It is out contention that improvement is hindered by a lack of theoretical understanding of their underlying mechanisms and dynamics.

Chatbots are the next big improvement in the era of conversational services. A chatbot is a virtual person who can carry out a conversation with a human about a certain subject, using interactive textual skills. Currently, there are many cloud-based chatbots services that are being developed and improved such as IBM Watson, well known for winning the quiz show “Jeopardy!” in 2011. Chatbots are based on a large amount of structured data. They contains many examples of questions that are associated to a specific intent which represents what the user wants to say. Those associations are currently being done by hand, and this project focuses on improving this data structuring using both supervised and unsupervised algorithms. A supervised reclassification using an improved Barycenter method reached 85% in precision and 75% in recall for a data set containing 2005 questions. Questions that did not match any intent were then clustered in an unsupervised way using a K-means algorithm that reached a purity of 0.5 for the optimal K chosen.
Chatbots är nästa stora förbättring i konversationstiden. En chatbot är en virtuell person som kan genomföra en konversation med en människa om ett visst ämne, med hjälp av interaktiva textkunskaper. För närvarande finns det många molnbaserade chatbots-tjänster som utvecklas och förbättras som IBM Watson, känt för att vinna quizshowen "Jeopardy!" 2011. Chatbots baseras på en stor mängd strukturerade data. De innehåller många exempel på frågor som är kopplade till en specifik avsikt som representerar vad användaren vill säga. Dessa föreningar görs för närvarande för hand, och detta projekt fokuserar på att förbättra denna datastrukturering med hjälp av både övervakade och oövervakade algoritmer. En övervakad omklassificering med hjälp av en förbättrad Barycenter-metod uppnådde 85 % i precision och 75 % i recall för en dataset innehållande 2005 frågorna. Frågorna som inte matchade någon avsikt blev sedan grupperade på ett oövervakad sätt med en K-medelalgoritm som nådde en renhet på 0,5 för den optimala K som valts.

A generic missile warner decision algorithm for airborne platforms with an emphasis on multiple classifier systems is proposed within the scope of this thesis. For developing the algorithm, simulation data are utilized. The simulation data are created in order to cover a wide range of real-life scenarios and for this purpose a scenario creation methodology is proposed. The scenarios are simulated by a generic missile warner simulator and tracked object data for each scenario are produced. Various feature extraction techniques are applied to the output data of the scenarios and feature sets are generated. Feature sets are examined by using various statistical methods. The performance of selected multiple classifier systems are evaluated for all feature sets and experimental results are presented.

Les signatures manuscrites sont des traits biométriques comportementaux caractérisés par une grande variabilité intra-classe. Les modèles de Markov cachés (MMCs) ont été utilisés avec succès en vérification hors-ligne des signatures manuscrites (VHS) en raison de la nature séquentielle et très variable de la signature. En particulier, la topologie gauche-droite des MMCs est très bien adaptée aux caractéristiques de l’écriture occidentale, dont les mouvements de la main sont principalement exécutés de la gauche vers la droite. Comme la plupart des classificateurs de type génératif, les MMCs requièrent une quantité importante de données d’entraînement pour atteindre un niveau de performance élevé en généralisation. Malheureusement, le nombre de signatures disponibles pour l’apprentissage des VHS est très limité en pratique. De plus, uniquement les faux aléatoires sont utilisés pour l’apprentissage des VHS qui doivent être en mesure de discriminer entre les signatures authentiques et les classes de faux aléatoires, les faux simples et les imitations. Ces deux dernières classes de faux ne sont pas disponibles lors de la phase d’apprentissage. Les approches proposées dans cette thèse reposent sur le concept des classificateurs multiples basés sur des MMCs exploités pour l’extraction de plusieurs niveaux de perception des signatures. Cette stratégie basée sur la génération d’un nombre très important de caractéristiques permet la mise en oeuvre de classificateurs dans les sous-espaces aléatoires, ce qui permet de s’affranchir du nombre limité de données disponibles pour l’entraînement. Une nouvelle approche pour la combinaison des classificateurs basée sur le principe des hypothèses multiples dans l’espace ROC est proposée. Une technique de réparation des courbes ROC permet de s’affranchir du nombre limité de signatures disponibles et surtout pour l’évaluation de la performance des systèmes biométriques. Une deuxième contribution importante est la proposition d’une architecture de classification hybride de type génératif-discriminatif. L’utilisation conjointe des MMCs pour l’extraction des caractéristiques et des machines à vecteurs de support (MVSs) pour la classification permet une meilleure représentation non seulement de la classe des signatures authentiques, mais aussi de la classe des imposteurs. L’approche proposée permet un apprentissage plus robuste que les approches MMCs conventionnelles lorsque le nombre d’échantillons disponibles est limité. La dernière contribution de cette thèse est la proposition de deux nouvelles stratégies pour la sélection dynamique (SD) d’ensembles de classificateurs. Les résultats obtenus sur les bases de signaturesmanuscrites PUCPR et GPDS, montrent que les stratégies proposées sont plus performantes que celles publiées dans la littérature pour la sélection dynamique ou statique des ensembles de classificateurs.

Dans cette thèse, nous proposons plusieurs nouvelles techniques pour la classification d'images hyperspectrales basées sur l'apprentissage d'ensemble. Le cadre proposé introduit des innovations importantes par rapport aux approches précédentes dans le même domaine, dont beaucoup sont basées principalement sur un algorithme individuel. Tout d'abord, nous proposons d'utiliser la Forêt de Rotation (Rotation Forest) avec différentes techiniques d'extraction de caractéristiques linéaire et nous comparons nos méthodes avec les approches d'ensemble traditionnelles, tels que Bagging, Boosting, Sous-espace Aléatoire et Forêts Aléatoires. Ensuite, l'intégration des machines à vecteurs de support (SVM) avec le cadre de sous-espace de rotation pour la classification de contexte est étudiée. SVM et sous-espace de rotation sont deux outils puissants pour la classification des données de grande dimension. C'est pourquoi, la combinaison de ces deux méthodes peut améliorer les performances de classification. Puis, nous étendons le travail de la Forêt de Rotation en intégrant la technique d'extraction de caractéristiques locales et l'information contextuelle spatiale avec un champ de Markov aléatoire (MRF) pour concevoir des méthodes spatio-spectrale robustes. Enfin, nous présentons un nouveau cadre général, ensemble de sous-espace aléatoire, pour former une série de classifieurs efficaces, y compris les arbres de décision et la machine d'apprentissage extrême (ELM), avec des profils multi-attributs étendus (EMaPS) pour la classification des données hyperspectrales. Six méthodes d'ensemble de sous-espace aléatoire, y compris les sous-espaces aléatoires avec les arbres de décision, Forêts Aléatoires (RF), la Forêt de Rotation (RoF), la Forêt de Rotation Aléatoires (Rorf), RS avec ELM (RSELM) et sous-espace de rotation avec ELM (RoELM), sont construits par multiples apprenants de base. L'efficacité des techniques proposées est illustrée par la comparaison avec des méthodes de l'état de l'art en utilisant des données hyperspectrales réelles dans de contextes différents
In this thesis, we propose several new techniques for the classification of hyperspectral remote sensing images based on multiple classifier system (MCS). Our proposed framework introduces significant innovations with regards to previous approaches in the same field, many of which are mainly based on an individual algorithm. First, we propose to use Rotation Forests with several linear feature extraction and compared them with the traditional ensemble approaches, such as Bagging, Boosting, Random subspace and Random Forest. Second, the integration of the support vector machines (SVM) with Rotation subspace framework for context classification is investigated. SVM and Rotation subspace are two powerful tools for high-dimensional data classification. Therefore, combining them can further improve the classification performance. Third, we extend the work of Rotation Forests by incorporating local feature extraction technique and spatial contextual information with Markov random Field (MRF) to design robust spatial-spectral methods. Finally, we presented a new general framework, Random subspace ensemble, to train series of effective classifiers, including decision trees and extreme learning machine (ELM), with extended multi-attribute profiles (EMAPs) for classifying hyperspectral data. Six RS ensemble methods, including Random subspace with DT (RSDT), Random Forest (RF), Rotation Forest (RoF), Rotation Random Forest (RoRF), RS with ELM (RSELM) and Rotation subspace with ELM (RoELM), are constructed by the multiple base learners. The effectiveness of the proposed techniques is illustrated by comparing with state-of-the-art methods by using real hyperspectral data sets with different contexts

Granular computing facilitates dealing with information by providing a theoretical framework to deal with information as granules at different levels of granularity (different levels of specificity/abstraction). It aims to provide an abstract explainable description of the data by forming granules that represent the features or the underlying structure of corresponding subsets of the data. In this thesis, a granular computing approach to the design of intelligent classification systems is proposed. The proposed approach is employed for different classification systems to investigate its efficiency. Fuzzy inference systems, neural networks, neuro-fuzzy systems and classifier ensembles are considered to evaluate the efficiency of the proposed approach. Each of the considered systems is designed using the proposed approach and classification performance is evaluated and compared to that of the standard system. The proposed approach is based on constructing information granules from data at multiple levels of granularity. The granulation process is performed using a modified fuzzy c-means algorithm that takes classification problem into account. Clustering is followed by a coarsening process that involves merging small clusters into large ones to form a lower granularity level. The resulted granules are used to build each of the considered binary classifiers in different settings and approaches. Granules produced by the proposed granulation method are used to build a fuzzy classifier for each granulation level or set of levels. The performance of the classifiers is evaluated using real life data sets and measured by two classification performance measures: accuracy and area under receiver operating characteristic curve. Experimental results show that fuzzy systems constructed using the proposed method achieved better classification performance. In addition, the proposed approach is used for the design of neural network classifiers. Resulted granules from one or more granulation levels are used to train the classifiers at different levels of specificity/abstraction. Using this approach, the classification problem is broken down into the modelling of classification rules represented by the information granules resulting in more interpretable system. Experimental results show that neural network classifiers trained using the proposed approach have better classification performance for most of the data sets. In a similar manner, the proposed approach is used for the training of neuro-fuzzy systems resulting in similar improvement in classification performance. Lastly, neural networks built using the proposed approach are used to construct a classifier ensemble. Information granules are used to generate and train the base classifiers. The final ensemble output is produced by a weighted sum combiner. Based on the experimental results, the proposed approach has improved the classification performance of the base classifiers for most of the data sets. Furthermore, a genetic algorithm is used to determine the combiner weights automatically.

This thesis investigates Learning Classifier System architectures for decision making in continuous-valued domains. The information contained in continuous-valued domains is not always conveniently expressed using the ternary representation typically used by Learning Classifier Systems and an interval-based representation is a natural choice. Two intervalbased representations recently proposed are analysed, together with their associated operators. Evidence of considerable representational and operator bias is found. A new interval-based representation is proposed that is more straightforward than the previous ones and its bias is analysed. Learning Classifier Systems are compared for online environments that consist of real-valued states and which require every action made by the agent to count towards its performance. Two Learning Classifier System architecture are considered , XCS and ZCS. An interval representation is used for the rule conditions and a roultte wh is used for action selection. The performance of these two Learning Classifier system architectures is investigated on a set of abstract environments with both deterministic and stochastic reward functions. Although XCS clearly delivers superior performance in the deterministic environments tested, the simple ZCS architectur is found to be robust and able to equal or exceed the performance of XCS in the stochastic environments tested, especially those with more demanding characteristics, Aspects of the algorithm and parameter set of ZCS are studied on problems with real-valued states and a Boolean action space. Increased performance is found to result from the use of an update algorithm based on that of NewBoole, an earlier strength-based Learning Classifier System. A new operator, specialize, is introduced and found to be effective in combatting over general classifiers. The modified algorithm and parameter set is tested on several variants of three real-valued test problems The resulting Learning Classifier System is applied to simulated Foreign Exchange trading using an experimental setup and data previously presented in the literature. Results show that a simple Learning Classifier System is able to achieve a positive excess return in simulated trading.

A fundamental problem in machine vision is the classifcation of objects which may have unknown position, orientation, or a combination of these and other transformations. The massive amount of data required to accurately form an appearance-based model of an object under all values of shift and rotation transformations has discouraged the incorporation of the combination of both transformations into a single model representation. This Master's Thesis documents the theory and implementation of a hierarchical classifier, named the Information Theoretic Decision Tree system, which has the demonstrated ability to form appearance-based models of objects which are shift and rotation invariant which can be searched with a great reduction in evaluations over a linear sequential search. Information theory is utilized to obtain a measure of information gain in a feature space recursive segmentation algorithm which positions hyperplanes to local information gain maxima. This is accomplished dynamically through a process of local optimization based on a conjugate gradient technique enveloped by a simulated annealing optimization loop. Several target model training strategies have been developed for shift and rotation invariance, notably the method of exemplar grouping, in which any combination of rotation and translation transformations of target object views can be simulated and folded into the appearance-based model. The decision tree structure target models produced as a result of this process effciently represent the voluminous training data, according rapid test-time classification of objects.

Many classification algorithms are designed on the assumption that the population of interest is stationary, i.e. it does not change over time. However, there are many real-world problems where this assumption is not appropriate. In this thesis, we develop a classifier for non-stationary populations which is based on a multiple logistic model for the conditional class probabilities and incorporates a linear combination of the outputs of a number of pre-determined component classifiers. The final classifier is able to adjust to changes in the population by sequential updating of the coefficients of the linear combination, which are the parameters of the model. The model we use is motivated by the relatively good classification performance which has been achieved by classification rules based on combining classifier outputs. However, in some cases such classifiers can also perform relatively poorly, and in general the mechanisms behind such results are little understood. For the model we propose, which is a generalisation of several existing models for stationary classification problems, we show there exists a simple relationship between the component classifiers which are used, the sign of the parameters and the decision boundaries of the final classifier. This relationship can be used to guide the choice of component classifiers, and helps with understanding the conditions necessary for the classifier to perform well. We compare several "on-line" algorithms for implementing the classification model, where the classifier is updated as new labelled observations become available. The predictive approach to classification is adopted, so each algorithm is based on updating the posterior distribution of the parameters as new information is received. Specifically, we compare a method which assumes the posterior distribution is Gaussian, a more general method developed for the class of Dynamic Generalised Linear Models, and a method based on a sequential Monte Carlo approximation of the posterior. The relationship between the model used for parameter evolution, the bias of the parameter estimates and the error of the classifier is explored.

EMG signal decomposition is the process of resolving a composite EMG signal into its constituent motor unit potential trains (classes) and it can be configured as a classification problem. An EMG signal detected by the tip of an inserted needle electrode is the superposition of the individual electrical contributions of the different motor units that are active, during a muscle contraction, and background interference.
This thesis addresses the process of EMG signal decomposition by developing an interactive classification system, which uses multiple classifier fusion techniques in order to achieve improved classification performance. The developed system combines heterogeneous sets of base classifier ensembles of different kinds and employs either a one level classifier fusion scheme or a hybrid classifier fusion approach.
The hybrid classifier fusion approach is applied as a two-stage combination process that uses a new aggregator module which consists of two combiners: the first at the abstract level of classifier fusion and the other at the measurement level of classifier fusion such that it uses both combiners in a complementary manner. Both combiners may be either data independent or the first combiner data independent and the second data dependent. For the purpose of experimentation, we used as first combiner the majority voting scheme, while we used as the second combiner one of the fixed combination rules behaving as a data independent combiner or the fuzzy integral with the lambda-fuzzy measure as an implicit data dependent combiner.
Once the set of motor unit potential trains are generated by the classifier fusion system, the firing pattern consistency statistics for each train are calculated to detect classification errors in an adaptive fashion. This firing pattern analysis allows the algorithm to modify the threshold of assertion required for assignment of a motor unit potential classification individually for each train based on an expectation of erroneous assignments.
The classifier ensembles consist of a set of different versions of the Certainty classifier, a set of classifiers based on the nearest neighbour decision rule: the fuzzy k-NN and the adaptive fuzzy k-NN classifiers, and a set of classifiers that use a correlation measure as an estimation of the degree of similarity between a pattern and a class template: the matched template filter classifiers and its adaptive counterpart. The base classifiers, besides being of different kinds, utilize different types of features and their performances were investigated using both real and simulated EMG signals of different complexities. The feature sets extracted include time-domain data, first- and second-order discrete derivative data, and wavelet-domain data.
Following the so-called overproduce and choose strategy to classifier ensemble combination, the developed system allows the construction of a large set of candidate base classifiers and then chooses, from the base classifiers pool, subsets of specified number of classifiers to form candidate classifier ensembles. The system then selects the classifier ensemble having the maximum degree of agreement by exploiting a diversity measure for designing classifier teams. The kappa statistic is used as the diversity measure to estimate the level of agreement between the base classifier outputs, i. e. , to measure the degree of decision similarity between the base classifiers. This mechanism of choosing the team's classifiers based on assessing the classifier agreement throughout all the trains and the unassigned category is applied during the one level classifier fusion scheme and the first combiner in the hybrid classifier fusion approach. For the second combiner in the hybrid classifier fusion approach, we choose team classifiers also based on kappa statistics but by assessing the classifiers agreement only across the unassigned category and choose those base classifiers having the minimum agreement.
Performance of the developed classifier fusion system, in both of its variants, i. e. , the one level scheme and the hybrid approach was evaluated using synthetic simulated signals of known properties and real signals and then compared it with the performance of the constituent base classifiers. Across the EMG signal data sets used, the hybrid approach had better average classification performance overall, specially in terms of reducing the number of classification errors.

Durant l'última dècada, els sistemes classificadors (LCS) d'estil Michigan - sistemes d'aprenentatge automàtic que combinen tècniques de repartiment de crèdit i algorismes genètics (AG) per evolucionar una població de classificadors online- han renascut. Juntament amb la formulació dels sistemes de primera generació, s'han produït avenços importants en (1) el disseny sistemàtic de nous LCS competents, (2) la seva aplicació en dominis rellevants i (3) el desenvolupament d'anàlisis teòriques. Malgrat aquests dissenys i aplicacions importants, encara hi ha reptes complexos que cal abordar per comprendre millor el funcionament dels LCS i per solucionar problemes del món real eficientment i escalable.
Aquesta tesi tracta dos reptes importants - compartits amb la comunitat d'aprenentatge automàtic - amb LCS d'estil Michigan: (1) aprenentatge en dominis que contenen classes estranyes i (2) evolució de models comprensibles on s'utilitzin mètodes de raonament similars als humans. L'aprenentatge de models precisos de classes estranyes és crític, doncs el coneixement clau sol quedar amagat en exemples d'aquestes, i la majoria de tècniques d'aprenentatge no són capaces de modelar la raresa amb precisió. La detecció de rareses sol ser complicat en aprenentatge online ja que el sistema d'aprenentatge rep un flux d'exemples i ha de detectar les rareses al vol. D'altra banda, l'evolució de models comprensibles és crucial en certs dominis com el mèdic, on l'expert acostuma a estar més interessat en obtenir una explicació intel·ligible de la predicció que en la predicció en si mateixa.
El treball present considera dos LCS d'estil Michigan com a punt de partida: l'XCS i l 'UCS. Es pren l'XCS com a primera referència ja que és l'LCS que ha tingut més influencia fins al moment. L'UCS hereta els components principals de l'XCS i els especialitza per aprenentatge supervisat. Tenint en compte que aquesta tesi especialment se centra en problemes de classificació, l'UCS també es considera en aquest estudi. La inclusió de l'UCS marca el primer objectiu de la tesi, sota el qual es revisen un conjunt de punts que van restar oberts en el disseny del sistema. A més, per il·lustrar les diferències claus entre l'XCS i l'UCS, es comparen ambdós sistemes sobre una bateria de problemes artificials de complexitat acotada.
L'estudi de com els LCS aprenen en dominis amb classes estranyes comença amb un estudi analític que descompon el problema en cinc elements crítics i deriva models per facetes per cadascun d'ells. Aquesta anàlisi s'usa com a eina per dissenyar guies de configuració que permeten que l'XCS i l'UCS solucionin problemes que prèviament no eren resolubles. A continuació, es comparen els dos LCS amb alguns dels sistemes d'aprenentatge amb més influencia en la comunitat d'aprenentatge automàtic sobre una col·lecció de problemes del món real que contenen classes estranyes. Els resultats indiquen que els dos LCS són els mètodes més robustos de la comparativa. Així mateix, es demostra experimentalment que remostrejar els conjunts d'entrenament amb l'objectiu d'eliminar la presencia de classes estranyes beneficia, en mitjana, el rendiment de les tècniques d'aprenentatge.
El repte de crear models més comprensibles i d'usar mecanismes de raonament que siguin similars als humans s'aborda mitjançant el disseny d'un nou LCS per aprenentatge supervisat que combina les capacitats d'avaluació de regles online, la robustesa mostrada pels AG en problemes complexos i la representació comprensible i mètodes de raonament fonamentats proporcionats per la lògica difusa. El nou LCS, anomenat Fuzzy-UCS, s'estudia en detall i es compara amb una bateria de mètodes d'aprenentatge. Els resultats de la comparativa demostren la competitivitat del Fuzzy-UCS en termes de precisió i intel·ligibilitat dels models evolucionats. Addicionalment, s'usa Fuzzy-UCS per extreure models de classificació acurats de grans volums de dades, exemplificant els avantatges de l'arquitectura d'aprenentatge online del Fuzzy-UCS.
En general, les observacions i avenços assolits en aquesta tesi contribueixen a augmentar la comprensió del funcionament dels LCS i en preparar aquests tipus de sistemes per afrontar problemes del món real de gran complexitat. Finalment, els resultats experimentals ressalten la robustesa i competitivitat dels LCS respecte a altres mètodes d'aprenentatge, encoratjant el seu ús per tractar nous problemes del món real.
Durante la última década, los sistemas clasificadores (LCS) de estilo Michigan - sistemas de aprendizaje automático que combinan técnicas de repartición de crédito y algoritmos genéticos (AG) para evolucionar una población de clasificadores online - han renacido. Juntamente con la formulación de los sistemas de primera generación, se han producido avances importantes en (1) el diseño sistemático de nuevos LCS competentes, (2) su aplicación en dominios relevantes y (3) el desarrollo de análisis teóricos. Pese a eso, aún existen retos complejos que deben ser abordados para comprender mejor el funcionamiento de los LCS y para solucionar problemas del mundo real escalable y eficientemente.
Esta tesis trata dos retos importantes - compartidos por la comunidad de aprendizaje automático - con LCS de estilo Michigan: (1) aprendizaje en dominios con clases raras y (2) evolución de modelos comprensibles donde se utilicen métodos de razonamiento similares a los humanos. El aprendizaje de modelos precisos de clases raras es crítico pues el conocimiento clave suele estar escondido en ejemplos de estas clases, y la mayoría de técnicas de aprendizaje no son capaces de modelar la rareza con precisión. El modelado de las rarezas acostumbra a ser más complejo en entornos de aprendizaje online, pues el sistema de aprendizaje recibe un flujo de ejemplos y debe detectar las rarezas al vuelo. La evolución de modelos comprensibles es crucial en ciertos dominios como el médico, donde el experto está más interesado en obtener una explicación inteligible de la predicción que en la predicción en sí misma.
El trabajo presente considera dos LCS de estilo Michigan como punto de partida: el XCS y el UCS. Se toma XCS como primera referencia debido a que es el LCS que ha tenido más influencia hasta el momento. UCS es un diseño reciente de LCS que hereda los componentes principales de XCS y los especializa para aprendizaje supervisado. Dado que esta tesis está especialmente centrada en problemas de clasificación automática, también se considera UCS en el estudio. La inclusión de UCS marca el primer objetivo de la tesis, bajo el cual se revisan un conjunto de aspectos que quedaron abiertos durante el diseño del sistema. Además, para ilustrar las diferencias claves entre XCS y UCS, se comparan ambos sistemas sobre una batería de problemas artificiales de complejidad acotada.
El estudio de cómo los LCS aprenden en dominios con clases raras empieza con un estudio analítico que descompone el problema en cinco elementos críticos y deriva modelos por facetas para cada uno de ellos. Este análisis se usa como herramienta para diseñar guías de configuración que permiten que XCS y UCS solucionen problemas que previamente no eran resolubles. A continuación, se comparan los dos LCS con algunos de los sistemas de aprendizaje de mayor influencia en la comunidad de aprendizaje automático sobre una colección de problemas del mundo real que contienen clases raras.
Los resultados indican que los dos LCS son los métodos más robustos de la comparativa. Además, se demuestra experimentalmente que remuestrear los conjuntos de entrenamiento con el objetivo de eliminar la presencia de clases raras beneficia, en promedio, el rendimiento de los métodos de aprendizaje automático incluidos en la comparativa.
El reto de crear modelos más comprensibles y usar mecanismos de razonamiento que sean similares a los humanos se aborda mediante el diseño de un nuevo LCS para aprendizaje supervisado que combina las capacidades de evaluación de reglas online, la robustez mostrada por los AG en problemas complejos y la representación comprensible y métodos de razonamiento proporcionados por la lógica difusa. El sistema que resulta de la combinación de estas ideas, llamado Fuzzy-UCS, se estudia en detalle y se compara con una batería de métodos de aprendizaje altamente reconocidos en el campo de aprendizaje automático. Los resultados de la comparativa demuestran la competitividad de Fuzzy-UCS en referencia a la precisión e inteligibilidad de los modelos evolucionados. Adicionalmente, se usa Fuzzy-UCS para extraer modelos de clasificación precisos de grandes volúmenes de datos, ejemplificando las ventajas de la arquitectura de aprendizaje online de Fuzzy-UCS.
En general, los avances y observaciones proporcionados en la tesis presente contribuyen a aumentar la comprensión del funcionamiento de los LCS y a preparar estos tipos de sistemas para afrontar problemas del mundo real de gran complejidad. Además, los resultados experimentales resaltan la robustez y competitividad de los LCS respecto a otros métodos de aprendizaje, alentando su uso para tratar nuevos problemas del mundo real.
During the last decade, Michigan-style learning classifier systems (LCSs) - genetic-based machine learning (GBML) methods that combine apportionment of credit techniques and genetic algorithms (GAs) to evolve a population of classifiers online - have been enjoying a renaissance. Together with the formulation of first generation systems, there have been crucial advances in (1) systematic design of new competent LCSs, (2) applications in important domains, and (3) theoretical analyses for design. Despite these successful designs and applications, there still remain difficult challenges that need to be addressed to increase our comprehension of how LCSs behave and to scalably and efficiently solve real-world problems.
The purpose of this thesis is to address two important challenges - shared by the machine learning community - with Michigan-style LCSs: (1) learning from domains that contain rare classes and (2) evolving highly legible models in which human-like reasoning mechanisms are employed. Extracting accurate models from rare classes is critical since the key, unperceptive knowledge usually resides in the rarities, and many traditional learning techniques are not able to model rarity accurately. Besides, these difficulties are increased in online learning, where the learner receives a stream of examples and has to detect rare classes on the fly. Evolving highly legible models is crucial in some domains such as medical diagnosis, in which human experts may be more interested in the explanation of the prediction than in the prediction itself.
The contributions of this thesis take two Michigan-style LCSs as starting point: the extended classifier system (XCS) and the supervised classifier system (UCS). XCS is taken as the first reference of this work since it is the most influential LCS. UCS is a recent LCS design that has inherited the main components of XCS and has specialized them for supervised learning. As this thesis is especially concerned with classification problems, UCS is also considered in this study. Since UCS is still a young system, for which there are several open issues that need further investigation, its learning architecture is first revised and updated. Moreover, to illustrate the key differences between XCS and UCS, the behavior of both systems is compared % and show that UCS converges quickly than XCS on a collection of boundedly difficult problems.
The study of learning from rare classes with LCSs starts with an analytical approach in which the problem is decomposed in five critical elements, and facetwise models are derived for each element. The analysis is used as a tool for designing configuration guidelines that enable XCS and UCS to solve problems that previously eluded solution. Thereafter, the two LCSs are compared with several highly-influential learners on a collection of real-world problems with rare classes, appearing as the two best techniques of the comparison. Moreover, re-sampling the training data set to eliminate the presence of rare classes is demonstrated to benefit, on average, the performance of LCSs.
The challenge of building more legible models and using human-like reasoning mechanisms is addressed with the design of a new LCS for supervised learning that combines the online evaluation capabilities of LCSs, the search robustness over complex spaces of GAs, and the legible knowledge representation and principled reasoning mechanisms of fuzzy logic. The system resulting from this crossbreeding of ideas, referred to as Fuzzy-UCS, is studied in detail and compared with several highly competent learning systems, demonstrating the competitiveness of the new architecture in terms of the accuracy and the interpretability of the evolved models. In addition, the benefits provided by the online architecture are exemplified by extracting accurate classification models from large data sets.
Overall, the advances and key insights provided in this thesis help advance our understanding of how LCSs work and prepare these types of systems to face increasingly difficult problems, which abound in current industrial and scientific applications. Furthermore, experimental results highlight the robustness and competitiveness of LCSs with respect to other machine learning techniques, which encourages their use to face new challenging real-world applications.

I seek an understanding of the dynamics: learning and evolution, of certain groups of artificially created agents - known here as trader-types - making decisions in a real stock market scenario. With this as the primary motivation, a specific problem has been devised and three different groups of agents have been modelled to learn, forecast and trade in a real stock market scenario given exogenously in the form of easily-obtained stock statistics such as various price moving averages, first difference in prices, volume ratios, etc. These artificial trader-types trade and learn simultaneously during - in most cases - a ten year period. They start with no prior knowledge about the market, i.e. they have no notion of what is a good or a bad approach to start with; all their market models are created randomly at the beginning of the period, with the idea that new models developed through experience will be formed and polished as time progresses. The life of such trader-types commences when they are given an initial wealth to trade over two assets (a risk less bond represented by the fixed interest rate given by the bank and a real risky stock) and ends in most cases after one decade. First, in this problem I try to explore whether it is feasible to represent with Learning Classifier Systems (LCS) some of the key elements that play a role in the decision-making process of real stock market traders when viewing it from an evolutionary framework. Specifically, two fundamental questions are addressed under this first and broad topic: Are the trader-types able to (i) evolve and (ii) behave in similar ways to human traders under the real market conditions described above? Here the work is concentrated in LCS as the learning approach and in viewing the agent as part of a process where adaptation to a partially understood market environment is a necessary element for survival to occur. Second, this thesis reports on a number of experiments where the forecasting performance of the adaptive agents is compared against the performance of the buy-andhold strategy, a trend-following strategy, a random strategy and finally against the bank investment over the same period of time at a fixed compound interest rate. To make the experiments as real as possible, agents also pay commissions on every trade. The results so far suggest that this is an excellent approach to make trading decisions in the stock market and that continual learning and adaptation not only play an important role but are also necessary elements in the decision-making process. Third, the concept of continual learning is addressed, and to show how the model constantly adapts to new market behaviour, additional experiments which include a number of real stocks are presented, followed by a discussion section.

The Learning Classifier System (LCS) and its descendant, XCS, are promising paradigms for machine learning design and implementation. Whereas LCS allows classifier payoff predictions to guide system performance, XCS focuses on payoff-prediction accuracy instead, allowing it to evolve "optimal" classifier sets in particular applications requiring rational thought. This research examines LCS and XCS performance in artificial situations with broad social/commercial parallels, created using the non-Markov Iterated Prisoner's Dilemma (IPD) game-playing scenario, where the setting is sometimes asymmetric and where irrationality sometimes pays. This research systematically perturbs a "conventional" IPD-playing LCS-based agent until it results in a full-fledged XCS-based agent, contrasting the simulated behavior of each LCS variant in terms of a number of performance measures. The intent is to examine the XCS paradigm to understand how it better copes with a given situation (if it does) than the LCS perturbations studied.Experiment results indicate that the majority of the architectural differences do have a significant effect on the agents' performance with respect to the performance measures used in this research. The results of these competitions indicate that while each architectural difference significantly affected its agent's performance, no single architectural difference could be credited as causing XCS's demonstrated superiority in evolving optimal populations. Instead, the data suggests that XCS's ability to evolve optimal populations in the multiplexer and IPD problem domains result from the combined and synergistic effects of multiple architectural differences.In addition, it is demonstrated that XCS is able to reliably evolve the Optimal Population [O] against the TFT opponent. This result supports Kovacs' Optimality Hypothesis in the IPD environment and is significant because it is the first demonstrated occurrence of this ability in an environment other than the multiplexer and Woods problem domains.It is therefore apparent that while XCS performs better than its LCS-based counterparts, its demonstrated superiority may not be attributed to a single architectural characteristic. Instead, XCS's ability to evolve optimal classifier populations in the multiplexer problem domain and in the IPD problem domain studied in this research results from the combined and synergistic effects of multiple architectural differences.

This thesis presents the case of dynamically and adaptively learning signatures for network intrusion detection using genetic based machine learning techniques. The two major criticisms of the signature based intrusion detection systems are their i) reliance on domain experts to handcraft intrusion signatures and ii) inability to detect previously unknown attacks or the attacks for which no signatures are available at the time. In this thesis, we present a biologically-inspired computational approach to address these two issues. This is done by adaptively learning maximally general rules, which are referred to as signatures, from network traffic through a supervised learning classifier system, UCS. The rules are learnt dynamically (i.e., using machine intelligence and without the requirement of a domain expert), and adaptively (i.e., as the data arrives without the need to relearn the complete model after presenting each data instance to the current model). Our approach is hybrid in that signatures for both intrusive and normal behaviours are learnt. The rule based profiling of normal behaviour allows for anomaly detection in that the events not matching any of the rules are considered potentially harmful and could be escalated for an action. We study the effect of key UCS parameters and operators on its performance and identify areas of improvement through this analysis. Several new heuristics are proposed that improve the effectiveness of UCS for the prediction of unseen and extremely rare intrusive activities. A signature extraction system is developed that adaptively retrieves signatures as they are discovered by UCS. The signature extraction algorithm is augmented by introducing novel subsumption operators that minimise overlap between signatures. Mechanisms are provided to adapt the main algorithm parameters to deal with online noisy and imbalanced class data. The performance of UCS, its variants and the signature extraction system is measured through standard evaluation metrics on a publicly available intrusion detection dataset provided during the 1999 KDD Cup intrusion detection competition. We show that the extended UCS significantly improves test accuracy and hit rate while significantly reducing the rate of false alarms and cost per example scores than the standard UCS. The results are competitive to the best systems participated in the competition in addition to our systems being online and incremental rule learners. The signature extraction system built on top of the extended UCS retrieves a magnitude smaller rule set than the base UCS learner without any significant performance loss. We extend the evaluation of our systems to real time network traffic which is captured from a university departmental server. A methodology is developed to build fully labelled intrusion detection dataset by mixing real background traffic with attacks simulated in a controlled environment. Tools are developed to pre-process the raw network data into feature vector format suitable for UCS and other related machine learning systems. We show the effectiveness of our feature set in detecting payload based attacks.

The field of agent control is concerned with the design and implementation of components that form an agent's control architecture. The interaction between these components determines how an agent?s sensor data and internal state combine to direct how the agent will act. Rule-based systems couple sensing and action in the form of condition-action rules and one class of such systems, learning classifier systems, has been extensively used in the design of adaptive agents. An adaptive agent explores an often unknown environment and uses its experience in its environment with the aim of improving its performance over time. The data an adaptive agent receives regarding the current state of its environment might be limited and ambiguous. In learning classifier systems, three different approaches to the problem of limited and ambiguous data from the environment have been: (1) to enable the agent to learn from its past experience, (2) to develop sequences of rules (in which rules may be linked implicitly or explicitly) and (3) multiagent LCSs. This thesis investigates the use of an adaptive agent?s past experience as a resource with which to perform a number of functions internal to the agent. These functions involve developing explicit sequences of rules, detecting and escaping from infinite loops, and firing and reinforcing rules. The first part of this thesis documents the design, implementation and evaluation of a control system that incorporates these functions. The control system is realised as a learning classifier system and is evaluated through experiments in a number of environments that provide limited and ambiguous stimuli. These experiments test the impact of explicit sequences of rules on the performance of a learning classifier system more thoroughly than previous research achieved. The use of explicit sequences of rules results in mixed performance in these environments and it is shown that while the use of these sequences in simple environments enables the rule space to be more effectively explored, in complex environments the behaviours developed by these sequences result in the agent stagnating more often in corners of the environment. Rather than endowing the rule-base with more rules, as in previous research with explicit sequences of rules, it is proposed that multiple interacting agents may enhance the exploration of the rule space in more complex environments. This approach is taken in the second part of this thesis, where the control system is used with multiple agents that interact by sharing rules. The aim of this interaction is to enhance the rule discovery process through cooperation between agents and thus improve the performance of the agents in their respective environments. It is shown that the benefit obtained from rule sharing is dependent on the environment and the type and amount of rule sharing used and that rule sharing is generally more beneficial in complex environments compared to simple environments. The properties of the rule-bases developed in these environments are examined in order to account for these results and it is shown that the type and amount of rule sharing most useful in each environment are dependent on these properties.

Submitted by João Arthur Martins (joao.arthur@ufpe.br) on 2015-03-12T17:38:41Z No. of bitstreams: 2 Dissertaçao Tiago de Lima.pdf: 1469834 bytes, checksum: 95a0326778b3d0f98bd35a7449d8b92f (MD5) license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5)
Approved for entry into archive by Daniella Sodre (daniella.sodre@ufpe.br) on 2015-03-13T14:23:38Z (GMT) No. of bitstreams: 2 Dissertaçao Tiago de Lima.pdf: 1469834 bytes, checksum: 95a0326778b3d0f98bd35a7449d8b92f (MD5) license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5)
Made available in DSpace on 2015-03-13T14:23:38Z (GMT). No. of bitstreams: 2 Dissertaçao Tiago de Lima.pdf: 1469834 bytes, checksum: 95a0326778b3d0f98bd35a7449d8b92f (MD5) license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) Previous issue date: 2013-02-26
In this dissertation, we present a methodology that aims the automatic construction of multi-classifiers systems based on the combination of selection and fusion. The presented method initially finds an optimum number of clusters for training data set and subsequently determines an ensemble for each cluster found. For model evaluation, the testing data set are submitted to clustering techniques and the nearest cluster to data input will emit a supervised response through its associated ensemble. Self-organizing maps were used in the clustering phase and multilayer perceptrons were used in the classification phase. Adaptive differential evolution has been used in this work in order to optimize the parameters and performance of the different techniques used in the classification and clustering phases. The proposed method, called SFJADE - Selection and Fusion (SF) via Adaptive Differential Evolution (JADE), has been tested on data compression of signals generated by artificial nose sensors and well-known classification problems, including cancer, card, diabetes, glass, heart, horse, soybean and thyroid. The experimental results have shown that the SFJADE method has a better performance than some literature methods while significantly outperforming most of the methods commonly used to construct Multi-Classifier Systems.
Nesta dissertação, nós apresentamos uma metodologia que almeja a construção automática de sistemas de múltiplos classificadores baseados em uma combinação de seleção e fusão. O método apresentado inicialmente encontra um número ótimo de grupos a partir do conjunto de treinamento e subsequentemente determina um comitê para cada grupo encontrado. Para avaliação do modelo, os dados de teste são submetidos à técnica de agrupamento e o grupo mais próximo do dado de entrada irá emitir uma resposta supervisionada por meio de seu comitê associado. Mapas Auto Organizáveis foi usado na fase de agrupamento e Perceptrons de múltiplas camadas na fase de classificação. Evolução Diferencial Adaptativa foi utilizada neste trabalho a fim de otimizar os parâmetros e desempenho das diferentes técnicas utilizadas nas fases de classificação e agrupamento de dados. O método proposto, chamado SFJADE – Selection and Fusion (SF) via Adaptive Differential Evolution (JADE), foi testado em dados gerados para sensores de um nariz artificial e problemas de referência em classificação de padrões, que são: cancer, card, diabetes, glass, heart, heartc e horse. Os resultados experimentais mostraram que SFJADE possui um melhor desempenho que alguns métodos da literatura, além de superar a maioria dos métodos geralmente usados para a construção de sistemas de múltiplos classificadores.

This thesis is concerned with the development of novel methods of classifying data that is not inherently clustered. The performance of these novel algorithms in finding classifications in this data will be compared with that of existing artificial intelligence methods.

Industrial Control Systems (ICSes), networked through conventional IT infrastructures, are vulnerable to attacks originating from network channels. Perimeter security techniques such as access control and firewalls have had limited success in mitigating such attacks due to the frequent updates required by standard computing platforms, third-party hardware and embedded process controllers. The high level of human-machine interaction also aids in circumventing perimeter defenses, making an ICS susceptible to attacks such as reprogramming of embedded controllers. The Stuxnet and Aurora attacks have demonstrated the vulnerabilities of ICS security and proved that these systems can be stealthily compromised. We present several run-time methods for preemptive intrusion detection in industrial control systems to enhance ICS security against reconfiguration and network attacks. A run-time prediction using a linear model of the physical plant and a neural-network based classifier trigger mechanism are proposed for preemptive detection of an attack. A standalone, safety preserving, optimal backup controller is implemented to ensure plant safety in case of an attack. The intrusion detection mechanism and the backup controller are instantiated in configurable hardware, making them invisible to operating software and ensuring their integrity in the presence of malicious software. Hardware implementation of our approach on an inverted pendulum system illustrates the performance of both techniques in the presence of reconfiguration and network attacks.
Master of Science

The ultimate goal of this research was to improve student performance by adjusting an Intelligent Tutoring System's (ITS) coaching strategy based on the student's mood. As a step toward this goal, this study evaluated the relationships between each student's mood variables (pleasure, arousal, dominance and mood intensity), the coaching strategy selected by the ITS and the student's performance. Outcomes included methods to increase the perception of the intelligent tutor to allow it to adapt coaching strategies (methods of instruction) to the student's affective needs to mitigate barriers to performance (e.g. negative affect) during the one-to-one tutoring process. The study evaluated whether the affective state (specifically mood) of the student moderated the student's interaction with the tutor and influenced performance. This research examined the relationships, interactions and influences of student mood in the selection of ITS coaching strategies to determine which strategies were more effective in terms of student performance given the student's mood, state (recent sleep time, previous knowledge and training, and interest level) and actions (e.g. mouse movement rate). Two coaching strategies were used in this study: Student-Requested Feedback (SRF) and Tutor-Initiated Feedback (TIF). The SRF coaching strategy provided feedback in the form of hints, questions, direction and support only when the student requested help. The TIF coaching strategy provided feedback (hints, questions, direction or support) at key junctures in the learning process when the student either made progress or failed to make progress in a timely fashion. The relationships between the coaching strategies, mood, performance and other variables of interest were considered in light of five hypotheses. At alpha = .05 and beta at least as great as .80, significant effects were limited in predicting performance. Highlighted findings include no significant differences in the mean performance due to coaching strategies, and only small effect sizes in predicting performance making the regression models developed not of practical significance. However, several variables including performance, energy level and mouse movement rates were significant, unobtrusive predictors of mood. Regression algorithms were developed using Arbuckle's (2008) Analysis of MOment Structures (AMOS) tool to compare the predicted performance for each strategy and then to choose the optimal strategy. A set of production rules were also developed to train a machine learning classifier using Witten & Frank's (2005) Waikato Environment for Knowledge Analysis (WEKA) toolset. The classifier was tested to determine its ability to recognize critical relationships and adjust coaching strategies to improve performance. This study found that the ability of the intelligent tutor to recognize key affective relationships contributes to improved performance. Study assumptions include a normal distribution of student mood variables, student state variables and student action variables and the equal mean performance of the two coaching strategy groups (student-requested feedback and tutor-initiated feedback ). These assumptions were substantiated in the study. Potential applications of this research are broad since its approach is application independent and could be used within ill-defined or very complex domains where judgment might be influenced by affect (e.g. study of the law, decisions involving risk of injury or death, negotiations or investment decisions). Recommendations for future research include evaluation of the temporal, as well as numerical, relationships of student mood, performance, actions and state variables.
Ph.D.
Other
Engineering and Computer Science
Modeling and Simulation PhD