Dans de nombreux problèmes en apprentissage supervisé, les entrées ont une structure de groupes connue et/ou clairement identifiable. Dans ce contexte, l'élaboration d'une règle de prédiction utilisant les groupes plutôt que les variables individuelles peut être plus pertinente tant au niveau des performances prédictives que de l'interprétation. L'objectif de la thèse est de développer des méthodes par arbres adaptées aux variables groupées. Nous proposons deux approches qui utilisent la structure groupée des variables pour construire des arbres de décisions. La première méthode permet de construire des arbres binaires en classification. Une coupure est définie par le choix d'un groupe et d'une combinaison linéaire des variables du dit groupe. La seconde approche, qui peut être utilisée en régression et en classification, construit un arbre non-binaire dans lequel chaque coupure est un arbre binaire. Ces deux approches construisent un arbre maximal qui est ensuite élagué. Nous proposons pour cela deux stratégies d'élagage dont une est une généralisation du minimal cost-complexity pruning. Les arbres de décision étant instables, nous introduisons une méthode de forêts aléatoires pour variables groupées. Outre l'aspect prédiction, ces méthodes peuvent aussi être utilisées pour faire de la sélection de groupes grâce à l'introduction d'indices d'importance des groupes. Ce travail est complété par une partie indépendante dans laquelle nous nous plaçons dans un cadre d'apprentissage non supervisé. Nous introduisons un nouvel algorithme de clustering. Sous des hypothèses classiques, nous obtenons des vitesses de convergence pour le risque de clustering de l'algorithme proposé. / In many problems in supervised learning, inputs have a known and/or obvious group structure. In this context, elaborating a prediction rule that takes into account the group structure can be more relevant than using an approach based only on the individual variables for both prediction accuracy and interpretation. The goal of this thesis is to develop some tree-based methods adapted to grouped variables. Here, we propose two new tree-based approaches which use the group structure to build decision trees. The first approach allows to build binary decision trees for classification problems. A split of a node is defined according to the choice of both a splitting group and a linear combination of the inputs belonging to the splitting group. The second method, which can be used for prediction problems in both regression and classification, builds a non-binary tree in which each split is a binary tree. These two approaches build a maximal tree which is next pruned. To this end, we propose two pruning strategies, one of which is a generalization of the minimal cost-complexity pruning algorithm. Since decisions trees are known to be unstable, we introduce a method of random forests that deals with groups of inputs. In addition to the prediction purpose, these new methods can be also use to perform group variable selection thanks to the introduction of some measures of group importance, This thesis work is supplemented by an independent part in which we consider the unsupervised framework. We introduce a new clustering algorithm. Under some classical regularity and sparsity assumptions, we obtain the rate of convergence of the clustering risk for the proposed alqorithm.

Semi-supervised learning (SSL) is a branch of machine learning focusing on using not only labeled data samples, but also unlabeled ones, in an effort to decrease the need for labeled data and thus allow using machine learning even when labeling large amounts of data would be too costly. Despite its quick development in the recent years, there are still issues left to be solved before it can be broadly deployed in practice. One of those issues is class distribution mismatch. It arises when the unlabeled data contains samples not belonging to the classes present in the labeled data. This confuses the training and can even lead to getting a classifier performing worse than a classifier trained on the available data in purely supervised fashion. We designed a filtration method called Unfavorable Data Filtering (UDF) which extracts important features from the data and then uses a similarity-based filter to filter the irrelevant data out according to those features. The filtering happens before any of the SSL training takes places, making UDF usable with any SSL algorithm. To judge its effectiveness, we performed many experiments, mainly on the CIFAR-10 dataset. We found out that UDF is capable of significantly improving the resulting accuracy when compared to not filtering the data, identified basic guidelines...

Background: Over the past 4-5 months, the Coronavirus has rapidly spread to all parts of the world. Research is continuing to find a cure for this disease while there is no exact reason for this outbreak. As the number of cases to test for Coronavirus is increasing rapidly day by day, it is impossible to test due to the time and cost factors. Over recent years, machine learning has turned very reliable in the medical field. Using machine learning to predict COVID-19 in patients will reduce the time delay for the results of the medical tests and modulate health workers to give proper medical treatment to them. Objectives: The main goal of this thesis is to develop a machine learning model that could predict whether a patient is suffering from COVID-19. To develop such a model, a literature study alongside an experiment is set to identify a suitable algorithm. To assess the features that impact the prediction model. Methods: A Systematic Literature Review is performed to identify the most suitable algorithms for the prediction model. Then through the findings of the literature study, an experimental model is developed for prediction of COVID-19 and to identify the features that impact the model. Results: A set of algorithms were identified from the Literature study that includes SVM (Support Vector Machines), RF (Random Forests), ANN (Artificial Neural Network), which are suitable for prediction. Performance evaluation is conducted between the chosen algorithms to identify the technique with the highest accuracy. Feature importance values are generated to identify their impact on the prediction. Conclusions: Prediction of COVID-19 by using Machine Learning could help increase the speed of disease identification resulting in reduced mortality rate. Analyzing the results obtained from experiments, Random Forest (RF) was identified to perform better compared to other algorithms.

Machine learning is a growing area of artificial intelligence that is widely used in our world today. Training machine learning models requires knowledge and computing power. Machine Learning as a Service (MLaaS) tries to solve these issues. By storing the datasets and using virtual computing instances in the cloud, one can create machine learning models without writing a single line of code. When selecting an MLaaS platform to use, the natural question of which one to use arises. This thesis conducts controlled experiments to compare the image classification capabilities of Microsoft Azure ML, Amazon Web Services SageMaker, and Google Cloud Platform Vertex AI. The prediction accuracy, training time, and cost will be measured with three different datasets. Some subjective comments about the user experience while conducting these experiments will also be provided. The results of these experiments will be used to make recommendations as to which MLaaS platform to use depending on which metric is most suitable. This thesis found that Microsoft Azure ML performed best in terms of prediction accuracy, and training cost, across all datasets. Amazon Web Services SageMaker had the shortest time to train but performed the worst in terms of accuracy and had trouble with two of the three datasets. Google Cloud Platform Vertex AI did achieve the second-bestprediction accuracy but was the most expensive platform by far as it had the largest time to train. It did, however, provide the smoothest user experience.Overall, Azure ML would be the platform of choice for image classification tasks after weighing together the results of the experiments as well as the subjective user experience.

Making predictions nowadays is of high importance for any company, whether small or large, as thanks to the possibility to analyze the data available, new market opportunities can be found, risks and costs can be reduced, among others. Machine learning algorithms for time series can be used for predicting future values of interest. However, choosing the appropriate algorithm and tuning its metaparameters require a great level of expertise. This creates an adoption barrier for small and medium enterprises which could not afford hiring a machine learning expert to their IT team. For these reasons, this project studies different possibilities to make good predictions based on machine learning algorithms, but without requiring great theoretical knowledge from the users. Moreover, a software package that implements the prediction process has been developed. The software is an ensemble method that first predicts a value taking into account different algorithms at the same time, and then it combines their results considering also the previous performance of each algorithm to obtain a final prediction of the value. Moreover, the solution proposed and implemented in this project can also predict according to a concrete objective (e.g., optimize the prediction, or do not exceed the real value) because not every prediction problem is subject to the same constraints. We have experimented and validated the implementation with three different cases. In all of them, a better performance has been obtained in comparison with each of the algorithms involved, reaching improvements of 45 to 95%.

Network attacks remain a constant threat to organizations around the globe. Intrusion detection systems provide a vital piece of the protection needed in order to fend off these attacks. Machine learning has become a popular method for developing new anomaly-based intrusion detection systems, and in recent years, deep learning has followed suit. Additionally, data fusion is often applied to intrusion detection systems in research, most often in the form of feature reduction, which can improve the accuracy and training times of classifiers. Another less common form of data fusion is decision fusion, where the outputs of multipe classifiers are fused into a more reliable result. Recent research has produced some contradictory results regarding the efficiency of traditional machine learning algorithms compared to deep learning algorithms. This study aims to investigate this problemand provide some clarity about the relative performance of a selection of classifier algorithms, namely artificial neural network, long short-term memory and random forest. Furthermore, two feature selection methods, namely correlation coefficient method and principal component analysis, as well as one decision fusion method in D-S evidence theory are tested. The majority of the feature selection methods fail to increase the accuracy of the implemented models, although the accuracy is not drastically reduced. Among the individual classifiers, random forest shows the best performance, obtaining an accuracy of 87,87%. Fusing the results with D-S evidence theory further improves this result, obtaining an accuracy of 88,56%, and proves particularly useful for reducing the number of false positives.

Negation is a complex phenomenon present in all human languages, allowing for the uniquely human capacities of denial, contradiction, misrepresentation, lying, and irony. It is in the first place a phenomenon of semantical opposition. Sentences containing negation are generally (a) less informative than affirmative ones, (b) morphosyntactically more marked—all languages have negative markers while only a few have affirmative markers, and (c) psychologically more complex and harder to process. Negation often conveys positive meaning. This meaning ranges from implicatures to entailments. In this dissertation, I develop a system to reveal the underlying positive interpretation of negation. I first identify which words are intended to be negated (i.e, the focus of negation) and second, I rewrite those tokens to generate an actual positive interpretation. I identify the focus of negation by scoring probable foci along a continuous scale. One of the obstacles to exploring foci scoring is that no public datasets exist for this task. Thus, to study this problem I create new corpora. The corpora contain verbal, nominal and adjectival negations and their potential positive interpretations along with their scores ranging from 1 to 5. Then, I use supervised learning models for scoring the focus of negation. In order to rewrite the focus of negation with its positive interpretation, I work with negations from Simple Wikipedia, automatically generate potential positive interpretations, and then collect manual annotations that effectively rewrite the negation in positive terms. This procedure yields positive interpretations for approximately 77% of negations, and the final corpus includes over 5,700 negations and over 5,900 positive interpretations. I then use sequence-to-sequence neural models and provide baseline results.

Background: Traffic Sign Recognition (TSR) is particularly useful for novice driversand self-driving cars. Driver Assistance Systems(DAS) involves automatic trafficsign recognition. Efficient classification of the traffic signs is required in DAS andunmanned vehicles for safe navigation. Convolutional Neural Networks(CNN) isknown for establishing promising results in the field of image classification, whichinspired us to employ this technique in our thesis. Computer vision is a process thatis used to understand the images and retrieve data from them. OpenCV is a Pythonlibrary used to detect traffic sign images in real-time. Objectives: This study deals with an experiment to build a CNN model which canclassify the traffic signs in real-time effectively using OpenCV. The model is builtwith low computational cost. The study also includes an experiment where variouscombinations of parameters are tuned to improve the model’s performance. Methods: The experimentation method involve building a CNN model based onmodified LeNet architecture with four convolutional layers, two max-pooling layersand two dense layers. The model is trained and tested with the German Traffic SignRecognition Benchmark (GTSRB) dataset. Parameter tuning with different combinationsof learning rate and epochs is done to improve the model’s performance.Later this model is used to classify the images introduced to the camera in real-time. Results: The graphs depicting the accuracy and loss of the model before and afterparameter tuning are presented. An experiment is done to classify the traffic signimage introduced to the camera by using the CNN model. High probability scoresare achieved during the process which is presented. Conclusions: The results show that the proposed model achieved 95% model accuracywith an optimum number of epochs, i.e., 30 and default optimum value oflearning rate, i.e., 0.001. High probabilities, i.e., above 75%, were achieved when themodel was tested using new real-time data.

Star Vault AB is a video game development company that has developed the video game Mortal Online. The company has stated that they believe that players new to the game repeatedly find themselves being lost in the game. The objective of this study is to evaluate whether or not an Artificial Neural Network can be used to evaluate when a player is lost in the game Mortal Online. This is done using the free open source library Fast Artifical Neural Network Library. People are invited to a data collection event where they play a tweaked version of the game to facilitate data collection. Players specify whether they are lost or not and the data collected is flagged accordingly. The collected data is then prepared with different parameters to be used when training multiple Artificial Neural Networks. When creating an Artificial Neural Network there exists several parameters which have an impact on its performance. Performance is defined as the balance of high prediction accuracy against low false positive rate. These parameters vary depending on the purpose of the Artificial Neural Network. A quantitative approach is followed where these parameters are varied to investigate which values result in the Artificial Neural Network which best identifies when a player is lost. The parameters are grouped into stages where all combinations of parameter values within each stage are evaluated to reduce the amount of Artificial Neural Networks which have to be trained, with the best performing parameters of each stage being used in subsequent stages. The result is a set of values for the parameters that are considered as ideal as possible. These parameter values are then altered one at a time to verify that they are ideal. The results show that a set of parameters exist which can optimize the Artificial Neural Network model to identify when a player is lost, however not with the high performance that was hoped for. It is theorized that the ambiguity of the word "lost" and the complexity of the game are critical to the low performance.

Deep learning has become ubiquitous in science and industry for classifying images or identifying patterns in data. The most widely used approach to training convolutional neural networks is supervised learning, which requires a large set of annotated data. To elude the high cost of collecting and annotating datasets, selfsupervised learning methods represent a promising way to learn the common functions of images and videos from large-scale unlabeled data without using humanannotated labels. This thesis provides the results of using self-supervised learning and explainable AI to localise objects in images from electron microscopes. The work used a synthetic geometric dataset and a synthetic pollen dataset. The classification was used as a pretext task. Different methods of explainable AI were applied: Grad-CAM and backpropagation-based approaches showed the lack of prospects; at the same time, the Extremal Perturbation function has shown efficiency. As a result of the downstream localisation task, the objects of interest were detected with competitive accuracy for one-class images. The advantages and limitations of the approach have been analysed. Directions for further work are proposed.