For many real-world applications, labeled data can be costly to obtain. Semi-supervised learning methods make use of substantially available unlabeled data along with few labeled samples. Most of the latest work on semi-supervised learning for image classification show performance on standard machine learning datasets like MNIST, SVHN, etc. In this work, we propose a convolutional adversarial autoencoder architecture for real-world data. We demonstrate the application of this architecture for semi-supervised object recognition. We show that our approach can learn from limited labeled data and outperform fully-supervised CNN baseline method by about 4% on real-world datasets. We also achieve competitive performance on the MNIST dataset compared to state-of-the-art semi-supervised learning techniques. To spur research in this direction, we compiled two real-world datasets: Internet (WIS) dataset and Real-world (RW) dataset which consists of more than 20K labeled samples each, comprising of small household objects belonging to ten classes. We also show a possible application of this method for online learning in robotics. / I de flesta verklighetsbaserade tillämpningar kan det vara kostsamt att erhålla märkt data. Inlärningsmetoder som är semi-övervakade använder sig oftast i stor utsträckning av omärkt data med stöd av en liten mängd märkt data. Mycket av det senaste arbetet inom semiövervakade inlärningsmetoder för bildklassificering visar prestanda på standardiserad maskininlärning så som MNIST, SVHN, och så vidare. I det här arbetet föreslår vi en convolutional adversarial autoencoder arkitektur för verklighetsbaserad data. Vi demonstrerar tillämpningen av denna arkitektur för semi-övervakad objektidentifiering och visar att vårt tillvägagångssätt kan lära sig av ett begränsat antal märkt data. Därmed överträffar vi den fullt övervakade CNN-baslinjemetoden med ca. 4% på verklighetsbaserade datauppsättningar. Vi uppnår även konkurrenskraftig prestanda på MNIST datauppsättningen jämfört med moderna semi-övervakade inlärningsmetoder. För att stimulera forskningen i den här riktningen, samlade vi två verklighetsbaserade datauppsättningar: Internet (WIS) och Real-world (RW) datauppsättningar, som består av mer än 20 000 märkta prov vardera, som utgörs av små hushållsobjekt tillhörandes tio klasser. Vi visar också en möjlig tillämpning av den här metoden för online-inlärning i robotik.

The selection of hospital once an ambulance has picked up its patient is today decided by the ambulance staff. This report describes a supervised machinelearning approach for predicting hospital selection. This is a multi-classclassification problem. The performance of random forest, logistic regression and neural network were compared to each other and to a baseline, namely the one rule-algorithm. The algorithms were applied to real world data from SOS-alarm, the company that operate Sweden’s emergency call services. Performance was measured with accuracy and f1-score. Random Forest got the best result followed by neural network. Logistic regression exhibited slightly inferior results but still performed far better than the baseline. The results point toward machine learning being a suitable method for learning the problem of hospital selection. / Beslutet om till vilket sjukhus en ambulans ska köra patienten till bestäms idag av ambulanspersonalen. Den här rapporten beskriver användandet av övervakad maskininlärning för att förutsåga detta beslut. Resultaten från algoritmerna slumpmässig skog, logistisk regression och neurala nätvärk jämförs med varanda och mot ett basvärde. Basvärdet erhölls med algorithmen en-regel. Algoritmerna applicerades på verklig data från SOS-alarm, Sveriges operatör för larmsamtal. Resultaten mättes med noggrannhet och f1-poäng. Slumpmässigskog visade bäst resultat följt av neurala nätverk. Logistisk regression uppvisade något sämre resultat men var fortfarande betydligt bättre än basvärdet. Resultaten pekar mot att det är lämpligt att använda maskininlärning för att lära sig att ta beslut om val av sjukhus.

Scraping is known to be difficult to detect and prevent, especially in the context of web APIs. It is in the interest of organisations that rely heavily on the content they provide through their web APIs to protect their content from scrapers. In this thesis, a machine learning approach towards detecting web API content scrapers is proposed. Three supervised machine learning algorithms were evaluated to see which would perform better on data from Spotify's web API. Data used to evaluate the classifiers consisted of aggregated HTTP request data that describes each application having sent HTTP requests to the web API over a span of two weeks. Two separate experiments were performed for each classifier, where the second experiment consisted of synthetic data for scrapers (the minority class) in addition to the original dataset. SMOTE was the algorithm used to perform oversampling in experiment two. The results show that Random Forest was the better classifier, with an MCC value of 0.692, without the use of synthetic data. For this particular problem, it is crucial that the classifier does not have a high false positive rate as legitimate usage of the web API should not be blocked. The Random Forest classifier has a low false positive rate and is therefore more favourable, and considered the strongest classifier out of the three examined. / Scraping är svårt att upptäcka och undvika, speciellt vad gäller att upptäcka applikationer som skrapar webb-APIer. Det finns ett särskilt intresse för organisationer, som är beroende av innehållet de tillhandahåller via sina webb-APIer, att skydda innehållet från applikationer som skrapar det. I denna avhandling föreslås ett tillvägagångssätt för att upptäcka dessa applikationer med hjälp av maskininlärning. Tre maskininlärningsalgoritmer utvärderades för att se vilka som skulle fungera bäst på data från Spotify's webb-API. Data som användes för att utvärdera dessa klassificerare bestod av aggregerade HTTP-request-data som beskriver varje applikation som har skickat HTTP-requests till webb-APIet under två veckors tid. Två separata experiment utfördes för varje klassificerare, där det andra experimentet var utökat med syntetisk data för applikationer som skrapar (minoritetsklassen) utöver det ursprungliga som användes i första experimentet. SMOTE var algoritmen som användes för att generera syntetisk data i experiment två. Resultaten visar att Random Forest var den bättre klassificeraren, med ett MCC-värde på 0,692, utan syntetisk data i det första experimentet. I detta fall är det viktigt att klassificeraren inte genererar många falska positiva resultat eftersom vanlig användning av ett web-API inte bör blockeras. Random Forest klassificeraren genererar få falska positiva resultat och är därför mer fördelaktig och anses vara den mest pålitliga klassificeraren av de tre undersökta.

On a high-level perspective, fashion is an art defined by fash- ion stylists and designers to express their thoughts and opinions. Lately, fashion have also been defined by digital publishers such as bloggers and online magazines. These digital publishers create fashion by curating and publishing content that is hopefully rel- evant and of high quality for their readers. Within this master’s thesis, fashion forecasting was investigated by applying supervised machine learning techniques. The problem was investigated by training classification learning models on a real world historical fashion dataset. The investigation has shown promising results, where fashion forecasting has been achieved with an average ac- curacy above 65 % . / På en abstrakt nivå definieras mode av stylister och designers.Dessa väljer att uttrycka sina tankar och åsikter genom att skapamode. På senare tid har mode också definierats av digitala förlagsom bloggare och onlinemagasin. Dessa digitala förlag definierarmode genom att skapa och publicera innehåll som förhoppningsvisär relevant och av hög kvalitet för sina läsare. I den här uppsatsen,undersöktes modeprognoser genom att använda sig av övervakademaskininlärningstekniker. Problemet undersöktes genom att läraklassificeringsinlärningsmodeller på ett verkligt historiskt datasetför mode. Undersökningen har visat lovande resultat där modeprognoserhar kunnat nås med en genomsnittlig noggrannhet över 65 %. / Maskininlärning, Förutspå Mode, Mode, Algoritmer, Klassificering

This thesis empirically studies transfer learning as a calibration framework for Convolutional Neural Network (CNN) based appearance-based gaze estimation models. A dataset of approximately 1,900,000 eyestripe images distributed over 1682 subjects is used to train and evaluate several gaze estimation models. Each model is initially trained on the training data resulting in generic gaze models. The models are subsequently calibrated for each test subject, using the subject's calibration data, by applying transfer learning through network fine-tuning on the final layers of the network. Transfer learning is observed to reduce the Euclidean distance error of the generic models within the range of 12-21%, which is in line with current state-of-the-art. The best performing calibrated model shows a mean error of 29.53mm and a median error of 22.77mm. However, calibrating heatmap output-based gaze estimation models decreases the performance over the generic models. It is concluded that transfer learning is a viable calibration framework for improving the performance of CNN-based appearance based gaze estimation models. / Detta examensarbete är en empirisk studie på överföringsträning som ramverk för kalibrering av neurala faltningsnätverks (CNN)-baserade bildbaserad blickapproximationsmodeller. En datamängd på omkring 1 900 000 ögonrandsbilder fördelat över 1682 personer används för att träna och bedöma flertalet blickapproximationsmodeller. Varje modell tränas inledningsvis på all träningsdata, vilket resulterar i generiska modeller. Modellerna kalibreras därefter för vardera testperson med testpersonens kalibreringsdata via överföringsträning genom anpassning av de sista lagren av nätverket. Med överföringsträning observeras en minskning av felet mätt som eukilidskt avstånd för de generiska modellerna inom 12-21%, vilket motsvarar de bästa nuvarande modellerna. För den bäst presterande kalibrerade modellen uppmäts medelfelet 29,53mm och medianfelet 22,77mm. Dock leder kalibrering av regionella sannolikhetsbaserade blickapproximationsmodeller till en försämring av prestanda jämfört med de generiska modellerna. Slutsatsen är att överföringsträning är en legitim kalibreringsansats för att förbättra prestanda hos CNN-baserade bildbaserad blickapproximationsmodeller.

Data is complex and has a variety of structures and formats. Modeling datasets is a core problem in modern artificial intelligence. Generative models are machine learning models, which model datasets with probability distributions. Deep generative models combine deep learning with probability theory, so that can model complicated datasets with flexible models. They have become one of the most popular models in machine learning, and have been applied to many problems. Normalizing flows are a novel class of deep generative models that allow efficient exact likelihood calculation, exact latent variable inference and sampling. They are constructed using functions whose inverse and Jacobian determinant can be efficiently computed. In this paper, we develop normalizing flow based generative models to model complex datasets. In general, data can be categorized to unlabeled data, labeled data, and weakly labeled data. We develop models for these three types of data, respectively. First, we develop Woodbury transformations, which are flow layers for general unsupervised normalizing flows, and can improve the flexibility and scalability of current flow based models. Woodbury transformations achieve efficient invertibility via Woodbury matrix identity and efficient determinant calculation via Sylvester's determinant identity. In contrast with other operations used in state-of-the-art normalizing flows, Woodbury transformations enable (1) high-dimensional interactions, (2) efficient sampling, and (3) efficient likelihood evaluation. Other similar operations, such as 1x1 convolutions, emerging convolutions, or periodic convolutions allow at most two of these three advantages. In our experiments on multiple image datasets, we find that Woodbury transformations allow learning of higher-likelihood models than other flow architectures while still enjoying their efficiency advantages. Second, we propose conditional Glow (c-Glow), a conditional generative flow for structured output learning, which is an advanced variant of supervised learning with structured labels. Traditional structured prediction models try to learn a conditional likelihood, i.e., p(y|x), to capture the relationship between the structured output y and the input features x. For many models, computing the likelihood is intractable. These models are therefore hard to train, requiring the use of surrogate objectives or variational inference to approximate likelihood. C-Glow benefits from the ability of flow-based models to compute p(y|x) exactly and efficiently. Learning with c-Glow does not require a surrogate objective or performing inference during training. Once trained, we can directly and efficiently generate conditional samples. We develop a sample-based prediction method, which can use this advantage to do efficient and effective inference. In our experiments, we test c-Glow on five different tasks. C-Glow outperforms the state-of-the-art baselines in some tasks and predicts comparable outputs in the other tasks. The results show that c-Glow is applicable to many different structured prediction problems. Third, we develop label learning flows (LLF), which is a general framework for weakly supervised learning problems. Our method is a generative model based on normalizing flows. The main idea of LLF is to optimize the conditional likelihoods of all possible labelings of the data within a constrained space defined by weak signals. We develop a training method for LLF that trains the conditional flow inversely and avoids estimating the labels. Once a model is trained, we can make predictions with a sampling algorithm. We apply LLF to three weakly supervised learning problems. Experiment results show that our method outperforms many state-of-the-art alternatives. Our research shows the advantages and versatility of normalizing flows. / Doctor of Philosophy / Data is now more affordable and accessible. At the same time, datasets are more and more complicated. Modeling data is a key problem in modern artificial intelligence and data analysis. Deep generative models combine deep learning and probability theory, and are now a major way to model complex datasets. In this dissertation, we focus on a novel class of deep generative model--normalizing flows. They are becoming popular because of their abilities to efficiently compute exact likelihood, infer exact latent variables, and draw samples. We develop flow-based generative models for different types of data, i.e., unlabeled data, labeled data, and weakly labeled data. First, we develop Woodbury transformations for unsupervised normalizing flows, which improve the flexibility and expressiveness of flow based models. Second, we develop conditional generative flows for an advanced supervised learning problem -- structured output learning, which removes the need of approximations, and surrogate objectives in traditional (deep) structured prediction models. Third, we develop label learning flows, which is a general framework for weakly supervised learning problems. Our research improves the performance of normalizing flows, and extend the applications of them to many supervised and weakly supervised problems.

This work is devoted to self-supervised representation learning (SSL). We consider both contrastive and non-contrastive methods and present a new loss function for SSL based on feature whitening. Our solution is conceptually simple and competitive with other methods. Self-supervised representations are beneficial for most areas of deep learning, and reinforcement learning is of particular interest because SSL can compensate for the sparsity of the training signal. We present two methods from this area. The first tackles the partial observability providing the agent with a history, represented with temporal alignment, and improves performance in most Atari environments. The second addresses the exploration problem. The method employs a world model of the SSL latent space, and the prediction error of this model indicates novel states required to explore. It shows strong performance on exploration-hard benchmarks, especially on the notorious Montezuma's Revenge. Finally, we consider the metric learning problem, which has much in common with SSL approaches. We present a new method based on hyperbolic embeddings, vision transformers and contrastive loss. We demonstrate the advantage of hyperbolic space over the widely used Euclidean space for metric learning. The method outperforms the current state-of-the-art by a significant margin.

Background : The heat load has an emergent role in the energy consumption of the heating system in buildings. The industry experts also have been constantly focusing on the heat load optimization techniques and in the recent years, numerous Machine Learning (ML) techniques have come into picture to resolve various tasks. Objectives : This study is mainly focused on to analyze the time-series hourly data and choose suitable Supervised Machine Learning approach among Multivariate Linear Regression (MLR), Support Vector Regression, and Multi-layer Perceptron (MLP) Regressor so as to predict heat demand for identifying the deviating behaviors and potentially faults. Methods : An experiment is performed and the method consists of imputing the missing values, extreme values and selection of six different feature sets. Cross validation on Multivariate Linear Regression, Support Vector Regression, and Multi-layer Perceptron Regressor was performed to find the best suitable algorithm. Finally the residuals of the best algorithm and the best feature set was used to find the fault using the calculation of studentized residuals. Because of the time-series based data in data set, regression based algorithms was the best suitable choice to work with such type of data that is continuous. The faults in the system were identified based on the studentized residuals that exceeds the threshold value of 3 are classified as fault. Results : Among the regression based algorithms, Multi-layer Perceptron Regressor resulted in Mean Absolute Error (MAE) of 1.77 and Mean Absolute Percentage Error (MAPE) 0.29% on the feature set 1. Multivariate Linear Regression shown Mean Absolute Error 1.83 and Mean Absolute Percentage Error 0.31% on feature set 1 that has relatively higher error for the metrics of Mean Absolute Error and Mean Absolute Percentage Error as comparing to Multi-layer Perceptron Regressor. Support Vector Regression (SVR) shown Mean Absolute Error 2.54 that is higher than that of both Multivariate Linear Regression and Multi-layer Perceptron Regressor, while theMean Absolute Percentage Error 0.24% that is similar to Multivariate Linear Regression and Multi-layer Perceptron Regressor on the feature set 1. So the best performing algorithm is Multi-layer Perceptron Regressor. The feature sets 4,5 and 6 which are super-sets of 1, 2 and 3 feature sets along with addition of outdoor temperature. These feature sets 4, 5 and 6 did not show much impact even after considering the outdoor temperature. From, the Table 5.1 the feature sets 1, 2 and 3 are comparitively better than feature sets 4, 5 and 6 for the metrics Mean Absolute Error and Mean Absolute Percentage Error.Finally on comparing the first three feature sets, the feature set 1 resulted in less error for all three algorithms as comparing to feature set 2 and feature set 3 that can be seen in Table 5.1. So the feature set 1 is the best feature set. Conclusions : Multi-layer Perceptron Regressor perfomed well on six different feature sets comparing with Multivariate Linear Regression and Support Vector Regression. The feature set 1 had shown Mean Absolute Error and Mean Absolute Percentage values relatively low than other feature sets. Therefore the feature set 1 was the best performing and the best suited algorithm was Multi-layer Perceptron Regressor. The Figure A.3 represents the flow of work done in the thesis.

Background: The train communication network is vulnerable to intrusion assaultsbecause of the openness of the ethernet communication protocol. Therefore, an intru-sion detection system must be incorporated into the train communication network.There are many algorithms available in Machine Learning(ML) to develop the Intru-sion Detection System(IDS). Majorly, depending on the accuracy and execution timeof the algorithm, it is decided as the best. Performance metrics like F1 score, preci-sion, recall, and support are compared to see how well the algorithm fits the modelwhile training. The following thesis will detect the anomalies in the Train ControlManagement System(TCMS) and then the comparison of various algorithms will beheld in order to declare the accurate algorithm. Objectives: In this thesis work, we aim to research anomaly detection in a train’sreal-time ethernet using an IDS. The main objectives of this thesis include per-forming Principal Component Analysis(PCA) and feature selection using RandomForest(RF) for simplifying the complexity of the dataset by reducing dimensionalityand extracting significant features. Followed by, choosing the most consistent algo-rithm for anomaly detection from the selected algorithms by evaluating performanceparameters, especially accuracy and execution time after training the models usingML algorithms. Method: This thesis necessitates one research methodology which is experimen-tation, to answer our research questions. For RQ1, experimentation will help usgain better insights into the dataset to extract valuable and essential features as apart of feature selection using RF and dimensionality reduction using PCA. RQ2also uses experimentation because it provides better accuracy and reliability. Afterpre-processing, the data will be used to train the algorithms and will be evaluatedusing various methods. Results: In this study, we have analysed data using EDA, reduced dimensionalityand feature selection using PCA and RF algorithm respectively. We used five su-pervised machine learning methods namely, Support Vector Machine(SVM), NaiveBayes, Decision Tree, K-nearest Neighbor(KNN), and Random Forest(RF). Aftertesting and utilizing the "KDDCup 1999" pre-processed dataset from the Universityof California Irvine(UCI) ML repository, Decision Tree model has been concludedas the best-performing algorithm with an accuracy of 98.89% in 0.098 seconds, incomparison to other models. Conclusions: Five models have been trained using the five ML techniques foranomaly detection using an IDS. We concluded that the decision tree trained modelhas optimal performance with an accuracy of 98.89% and time of 0.098 seconds

<p>Smart and connected technology solutions have emerged as a promising way to enhance EMS services, particularly in areas where access to professional services is limited. However, a significant challenge for improving their implementation is determining which technologies to use and how they will change current logistic operations to enhance service efficiencies and expand access to care. In this context, this thesis explores opportunities for the smart and connected technology solutions.</p> <p>The first study explores the use of medically trained volunteers in the community, known as Citizen Responders (CRs). These individuals can be quickly notified of an EMS request upon its arrival via a mobile alert receiver, which allows them to provide timely and potentially life-saving assistance before an ambulance arrives. However, traditional EMS logistic decision platforms are not equipped to effectively leverage the sharing of the real-time CR information enabled by connected technologies, such as their location and availability. To improve coordination between CRs and ambulances, this study proposes two decision tools that incorporate real-time CR information: one for redeploying ambulances after they complete service and another for dispatching ambulances in response to calls. The redeployment procedure uses mixed-integer linear programming (MILP) to maximize patient survival, while the dispatch procedure enhances a locally optimal dispatch procedure by integrating real-time CR information for priority-differentiated emergencies.</p> <p>In the second study, a third decision tool was developed to take advantage of the increasing availability of feature information provided by connected technologies: an AI-enabled dispatch rule recommendation model that is more usable for dispatchers than black-box decision models. This is a model based on supervised learning that outputs a “promising” metric-based dispatch rule for the human decision-maker. The model maintains the usability of rules while enhancing the system’s performance and alleviating the cognitive burden of dispatchers. A set of experiments were performed on a self-developed simulator to assess the performance of all the decision tools. The findings suggest they have the potential to significantly enhance the EMS system performance. </p>