Learning Preferences with Kernel-Based Methods

Learning of preference relations has recently received significant attention in machine learning community. It is closely related to the classification and regression analysis and can be reduced to these tasks. However, preference learning involves prediction of ordering of the data points rather than prediction of a single numerical value as in case of regression or a class label as in case of classification. Therefore, studying preference relations within a separate framework facilitates not only better theoretical understanding of the problem, but also motivates development of the efficient algorithms for the task. Preference learning has many applications in domains such as information retrieval, bioinformatics, natural language processing, etc. For example, algorithms that learn to rank are frequently used in search engines for ordering documents retrieved by the query. Preference learning methods have been also applied to collaborative filtering problems for predicting individual customer choices from the vast amount of user generated feedback. In this thesis we propose several algorithms for learning preference relations. These algorithms stem from well founded and robust class of regularized least-squares methods and have many attractive computational properties. In order to improve the performance of our methods, we introduce several non-linear kernel functions. Thus, contribution of this thesis is twofold: kernel functions for structured data that are used to take advantage of various non-vectorial data representations and the preference learning algorithms that are suitable for different tasks, namely efficient learning of preference relations, learning with large amount of training data, and semi-supervised preference learning. Proposed kernel-based algorithms and kernels are applied to the parse ranking task in natural language processing, document ranking in information retrieval, and remote homology detection in bioinformatics domain. Training of kernel-based ranking algorithms can be infeasible when the size of the training set is large. This problem is addressed by proposing a preference learning algorithm whose computation complexity scales linearly with the number of training data points. We also introduce sparse approximation of the algorithm that can be efficiently trained with large amount of data. For situations when small amount of labeled data but a large amount of unlabeled data is available, we propose a co-regularized preference learning algorithm. To conclude, the methods presented in this thesis address not only the problem of the efficient training of the algorithms but also fast regularization parameter selection, multiple output prediction, and cross-validation. Furthermore, proposed algorithms lead to notably better performance in many preference learning tasks considered.

Kernel-Based Ranking. Methods for Learning and Performance Estimation

Machine learning provides tools for automated construction of predictive models in data intensive areas of engineering and science. The family of regularized kernel methods have in the recent years become one of the mainstream approaches to machine learning, due to a number of advantages the methods share. The approach provides theoretically well-founded solutions to the problems of under- and overfitting, allows learning from structured data, and has been empirically demonstrated to yield high predictive performance on a wide range of application domains. Historically, the problems of classification and regression have gained the majority of attention in the field. In this thesis we focus on another type of learning problem, that of learning to rank. In learning to rank, the aim is from a set of past observations to learn a ranking function that can order new objects according to how well they match some underlying criterion of goodness. As an important special case of the setting, we can recover the bipartite ranking problem, corresponding to maximizing the area under the ROC curve (AUC) in binary classification. Ranking applications appear in a large variety of settings, examples encountered in this thesis include document retrieval in web search, recommender systems, information extraction and automated parsing of natural language. We consider the pairwise approach to learning to rank, where ranking models are learned by minimizing the expected probability of ranking any two randomly drawn test examples incorrectly. The development of computationally efficient kernel methods, based on this approach, has in the past proven to be challenging. Moreover, it is not clear what techniques for estimating the predictive performance of learned models are the most reliable in the ranking setting, and how the techniques can be implemented efficiently. The contributions of this thesis are as follows. First, we develop RankRLS, a computationally efficient kernel method for learning to rank, that is based on minimizing a regularized pairwise least-squares loss. In addition to training methods, we introduce a variety of algorithms for tasks such as model selection, multi-output learning, and cross-validation, based on computational shortcuts from matrix algebra. Second, we improve the fastest known training method for the linear version of the RankSVM algorithm, which is one of the most well established methods for learning to rank. Third, we study the combination of the empirical kernel map and reduced set approximation, which allows the large-scale training of kernel machines using linear solvers, and propose computationally efficient solutions to cross-validation when using the approach. Next, we explore the problem of reliable cross-validation when using AUC as a performance criterion, through an extensive simulation study. We demonstrate that the proposed leave-pair-out cross-validation approach leads to more reliable performance estimation than commonly used alternative approaches. Finally, we present a case study on applying machine learning to information extraction from biomedical literature, which combines several of the approaches considered in the thesis. The thesis is divided into two parts. Part I provides the background for the research work and summarizes the most central results, Part II consists of the five original research articles that are the main contribution of this thesis.

Comparison of classification ability of hyperball algorithms to neural network and k-nearest neighbour algorithms

The main focus of this thesis is to evaluate and compare Hyperbalilearning algorithm (HBL) to other learning algorithms. In this work HBL is compared to feed forward artificial neural networks using back propagation learning, K-nearest neighbor and 103 algorithms. In order to evaluate the similarity of these algorithms, we carried out three experiments using nine benchmark data sets from UCI machine learning repository. The first experiment compares HBL to other algorithms when sample size of dataset is changing. The second experiment compares HBL to other algorithms when dimensionality of data changes. The last experiment compares HBL to other algorithms according to the level of agreement to data target values. Our observations in general showed, considering classification accuracy as a measure, HBL is performing as good as most ANn variants. Additionally, we also deduced that HBL.:s classification accuracy outperforms 103's and K-nearest neighbour's for the selected data sets.

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.

Semi-online neural-Q_leaming for real-time robot learning

Reinforcement learning (RL) is a very suitable technique for robot learning, as it can learn in unknown environments and in real-time computation. The main difficulties in adapting classic RL algorithms to robotic systems are the generalization problem and the correct observation of the Markovian state. This paper attempts to solve the generalization problem by proposing the semi-online neural-Q_learning algorithm (SONQL). The algorithm uses the classic Q_learning technique with two modifications. First, a neural network (NN) approximates the Q_function allowing the use of continuous states and actions. Second, a database of the most representative learning samples accelerates and stabilizes the convergence. The term semi-online is referred to the fact that the algorithm uses the current but also past learning samples. However, the algorithm is able to learn in real-time while the robot is interacting with the environment. The paper shows simulated results with the "mountain-car" benchmark and, also, real results with an underwater robot in a target following behavior

Increasing detection rate of user-to-root attacks using genetic algorithms

An extensive set of machine learning and pattern classification techniques trained and tested on KDD dataset failed in detecting most of the user-to-root attacks. This paper aims to provide an approach for mitigating negative aspects of the mentioned dataset, which led to low detection rates. Genetic algorithm is employed to implement rules for detecting various types of attacks. Rules are formed of the features of the dataset identified as the most important ones for each attack type. In this way we introduce high level of generality and thus achieve high detection rates, but also gain high reduction of the system training time. Thenceforth we re-check the decision of the user-to- root rules with the rules that detect other types of attacks. In this way we decrease the false-positive rate. The model was verified on KDD 99, demonstrating higher detection rates than those reported by the state- of-the-art while maintaining low false-positive rate.

A hybrid approach to learn with imbalanced classes using evolutionary algorithms

There is an increasing interest in the application of Evolutionary Algorithms (EAs) to induce classification rules. This hybrid approach can benefit areas where classical methods for rule induction have not been very successful. One example is the induction of classification rules in imbalanced domains. Imbalanced data occur when one or more classes heavily outnumber other classes. Frequently, classical machine learning (ML) classifiers are not able to learn in the presence of imbalanced data sets, inducing classification models that always predict the most numerous classes. In this work, we propose a novel hybrid approach to deal with this problem. We create several balanced data sets with all minority class cases and a random sample of majority class cases. These balanced data sets are fed to classical ML systems that produce rule sets. The rule sets are combined creating a pool of rules and an EA is used to build a classifier from this pool of rules. This hybrid approach has some advantages over undersampling, since it reduces the amount of discarded information, and some advantages over oversampling, since it avoids overfitting. The proposed approach was experimentally analysed and the experimental results show an improvement in the classification performance measured as the area under the receiver operating characteristics (ROC) curve.

Particle Competition and Cooperation in Networks for Semi-Supervised Learning

Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP)

Particle Competition and Cooperation in Networks for Semi-Supervised Learning with Concept Drift

Concept drift is a problem of increasing importance in machine learning and data mining. Data sets under analysis are no longer only static databases, but also data streams in which concepts and data distributions may not be stable over time. However, most learning algorithms produced so far are based on the assumption that data comes from a fixed distribution, so they are not suitable to handle concept drifts. Moreover, some concept drifts applications requires fast response, which means an algorithm must always be (re) trained with the latest available data. But the process of labeling data is usually expensive and/or time consuming when compared to unlabeled data acquisition, thus only a small fraction of the incoming data may be effectively labeled. Semi-supervised learning methods may help in this scenario, as they use both labeled and unlabeled data in the training process. However, most of them are also based on the assumption that the data is static. Therefore, semi-supervised learning with concept drifts is still an open challenge in machine learning. Recently, a particle competition and cooperation approach was used to realize graph-based semi-supervised learning from static data. In this paper, we extend that approach to handle data streams and concept drift. The result is a passive algorithm using a single classifier, which naturally adapts to concept changes, without any explicit drift detection mechanism. Its built-in mechanisms provide a natural way of learning from new data, gradually forgetting older knowledge as older labeled data items became less influent on the classification of newer data items. Some computer simulation are presented, showing the effectiveness of the proposed method.

Active Learning to Maximize Area Under the ROC Curve

In active learning, a machine learning algorithmis given an unlabeled set of examples U, and is allowed to request labels for a relatively small subset of U to use for training. The goal is then to judiciously choose which examples in U to have labeled in order to optimize some performance criterion, e.g. classification accuracy. We study how active learning affects AUC. We examine two existing algorithms from the literature and present our own active learning algorithms designed to maximize the AUC of the hypothesis. One of our algorithms was consistently the top performer, and Closest Sampling from the literature often came in second behind it. When good posterior probability estimates were available, our heuristics were by far the best.

Particle Competition and Cooperation in Networks for Semi-Supervised Learning

Semi-supervised learning is one of the important topics in machine learning, concerning with pattern classification where only a small subset of data is labeled. In this paper, a new network-based (or graph-based) semi-supervised classification model is proposed. It employs a combined random-greedy walk of particles, with competition and cooperation mechanisms, to propagate class labels to the whole network. Due to the competition mechanism, the proposed model has a local label spreading fashion, i.e., each particle only visits a portion of nodes potentially belonging to it, while it is not allowed to visit those nodes definitely occupied by particles of other classes. In this way, a "divide-and-conquer" effect is naturally embedded in the model. As a result, the proposed model can achieve a good classification rate while exhibiting low computational complexity order in comparison to other network-based semi-supervised algorithms. Computer simulations carried out for synthetic and real-world data sets provide a numeric quantification of the performance of the method.

Network-based stochastic semisupervised learning

Semisupervised learning is a machine learning approach that is able to employ both labeled and unlabeled samples in the training process. In this paper, we propose a semisupervised data classification model based on a combined random-preferential walk of particles in a network (graph) constructed from the input dataset. The particles of the same class cooperate among themselves, while the particles of different classes compete with each other to propagate class labels to the whole network. A rigorous model definition is provided via a nonlinear stochastic dynamical system and a mathematical analysis of its behavior is carried out. A numerical validation presented in this paper confirms the theoretical predictions. An interesting feature brought by the competitive-cooperative mechanism is that the proposed model can achieve good classification rates while exhibiting low computational complexity order in comparison to other network-based semisupervised algorithms. Computer simulations conducted on synthetic and real-world datasets reveal the effectiveness of the model.

Structure learning of graphical models for task-oriented robot grasping

In the collective imaginaries a robot is a human like machine as any androids in science fiction. However the type of robots that you will encounter most frequently are machinery that do work that is too dangerous, boring or onerous. Most of the robots in the world are of this type. They can be found in auto, medical, manufacturing and space industries. Therefore a robot is a system that contains sensors, control systems, manipulators, power supplies and software all working together to perform a task. The development and use of such a system is an active area of research and one of the main problems is the development of interaction skills with the surrounding environment, which include the ability to grasp objects. To perform this task the robot needs to sense the environment and acquire the object informations, physical attributes that may influence a grasp. Humans can solve this grasping problem easily due to their past experiences, that is why many researchers are approaching it from a machine learning perspective finding grasp of an object using information of already known objects. But humans can select the best grasp amongst a vast repertoire not only considering the physical attributes of the object to grasp but even to obtain a certain effect. This is why in our case the study in the area of robot manipulation is focused on grasping and integrating symbolic tasks with data gained through sensors. The learning model is based on Bayesian Network to encode the statistical dependencies between the data collected by the sensors and the symbolic task. This data representation has several advantages. It allows to take into account the uncertainty of the real world, allowing to deal with sensor noise, encodes notion of causality and provides an unified network for learning. Since the network is actually implemented and based on the human expert knowledge, it is very interesting to implement an automated method to learn the structure as in the future more tasks and object features can be introduced and a complex network design based only on human expert knowledge can become unreliable. Since structure learning algorithms presents some weaknesses, the goal of this thesis is to analyze real data used in the network modeled by the human expert, implement a feasible structure learning approach and compare the results with the network designed by the expert in order to possibly enhance it.

Learning with Kernels on Graphs: DAG-based kernels, data streams and RNA function prediction.

In many application domains data can be naturally represented as graphs. When the application of analytical solutions for a given problem is unfeasible, machine learning techniques could be a viable way to solve the problem. Classical machine learning techniques are defined for data represented in a vectorial form. Recently some of them have been extended to deal directly with structured data. Among those techniques, kernel methods have shown promising results both from the computational complexity and the predictive performance point of view. Kernel methods allow to avoid an explicit mapping in a vectorial form relying on kernel functions, which informally are functions calculating a similarity measure between two entities. However, the definition of good kernels for graphs is a challenging problem because of the difficulty to find a good tradeoff between computational complexity and expressiveness. Another problem we face is learning on data streams, where a potentially unbounded sequence of data is generated by some sources. There are three main contributions in this thesis. The first contribution is the definition of a new family of kernels for graphs based on Directed Acyclic Graphs (DAGs). We analyzed two kernels from this family, achieving state-of-the-art results from both the computational and the classification point of view on real-world datasets. The second contribution consists in making the application of learning algorithms for streams of graphs feasible. Moreover,we defined a principled way for the memory management. The third contribution is the application of machine learning techniques for structured data to non-coding RNA function prediction. In this setting, the secondary structure is thought to carry relevant information. However, existing methods considering the secondary structure have prohibitively high computational complexity. We propose to apply kernel methods on this domain, obtaining state-of-the-art results.

Orthogonal MCMC algorithms

Monte Carlo (MC) methods are widely used in signal processing, machine learning and stochastic optimization. A well-known class of MC methods are Markov Chain Monte Carlo (MCMC) algorithms. In this work, we introduce a novel parallel interacting MCMC scheme, where the parallel chains share information using another MCMC technique working on the entire population of current states. These parallel ?vertical? chains are led by random-walk proposals, whereas the ?horizontal? MCMC uses a independent proposal, which can be easily adapted by making use of all the generated samples. Numerical results show the advantages of the proposed sampling scheme in terms of mean absolute error, as well as robustness w.r.t. to initial values and parameter choice.