abstract: The widespread adoption of computer vision models is often constrained by the issue of domain mismatch. Models that are trained with data belonging to one distribution, perform poorly when tested with data from a different distribution. Variations in vision based data can be attributed to the following reasons, viz., differences in image quality (resolution, brightness, occlusion and color), changes in camera perspective, dissimilar backgrounds and an inherent diversity of the samples themselves. Machine learning techniques like transfer learning are employed to adapt computational models across distributions. Domain adaptation is a special case of transfer learning, where knowledge from a source domain is transferred to a target domain in the form of learned models and efficient feature representations. The dissertation outlines novel domain adaptation approaches across different feature spaces; (i) a linear Support Vector Machine model for domain alignment; (ii) a nonlinear kernel based approach that embeds domain-aligned data for enhanced classification; (iii) a hierarchical model implemented using deep learning, that estimates domain-aligned hash values for the source and target data, and (iv) a proposal for a feature selection technique to reduce cross-domain disparity. These adaptation procedures are tested and validated across a range of computer vision applications like object classification, facial expression recognition, digit recognition, and activity recognition. The dissertation also provides a unique perspective of domain adaptation literature from the point-of-view of linear, nonlinear and hierarchical feature spaces. The dissertation concludes with a discussion on the future directions for research that highlight the role of domain adaptation in an era of rapid advancements in artificial intelligence. / Dissertation/Thesis / Doctoral Dissertation Computer Science 2017

abstract: Machine learning models convert raw data in the form of video, images, audio, text, etc. into feature representations that are convenient for computational process- ing. Deep neural networks have proven to be very efficient feature extractors for a variety of machine learning tasks. Generative models based on deep neural networks introduce constraints on the feature space to learn transferable and disentangled rep- resentations. Transferable feature representations help in training machine learning models that are robust across different distributions of data. For example, with the application of transferable features in domain adaptation, models trained on a source distribution can be applied to a data from a target distribution even though the dis- tributions may be different. In style transfer and image-to-image translation, disen- tangled representations allow for the separation of style and content when translating images. This thesis examines learning transferable data representations in novel deep gen- erative models. The Semi-Supervised Adversarial Translator (SAT) utilizes adversar- ial methods and cross-domain weight sharing in a neural network to extract trans- ferable representations. These transferable interpretations can then be decoded into the original image or a similar image in another domain. The Explicit Disentangling Network (EDN) utilizes generative methods to disentangle images into their core at- tributes and then segments sets of related attributes. The EDN can separate these attributes by controlling the ow of information using a novel combination of losses and network architecture. This separation of attributes allows precise modi_cations to speci_c components of the data representation, boosting the performance of ma- chine learning tasks. The effectiveness of these models is evaluated across domain adaptation, style transfer, and image-to-image translation tasks. / Dissertation/Thesis / Masters Thesis Computer Science 2018

Deep understanding of urban mobility is of great significance for many real-world applications, such as urban traffic management and autonomous driving. This thesis develops deep learning methodologies to extract vehicle counts from streaming realtime video captured by multiple low resolution web cameras and construct maps of traffic density in a city environment; in particular, we focus on cameras installed in the Manhattan borough of NYC. The large-scale videos from these web cameras have low spatial and temporal resolution, high occlusion, large perspective, and variable environment conditions, making most existing methods to lose their efficacy. To overcome these challenges, the thesis develops several techniques: 1. a block-level regression model with a rank constraint to map the dense image feature into vehicle densities; 2. a deep multi-task learning framework based on fully convolutional neural networks to jointly learn vehicle density and vehicle count; 3. deep spatio-temporal networks for vehicle counting to incorporate temporal information of the traffic flow; and 4. multi-source domain adaptation mechanisms with adversarial learning to adapt the deep counting model to multiple cameras. To train and validate the proposed system, we have collected a largescale webcam traffic dataset CityCam that contains 60 million frames from 212 webcams installed in key intersections of NYC. Of there, 60; 000 frames have been annotated with rich information, leading to about 900; 000 annotated objects. To the best of our knowledge, it is the first and largest webcam traffic dataset with such large number of elaborate annotations. The proposed methods are integrated into the CityScapeEye system that has been extensively evaluated and compared to existing techniques on different counting tasks and datasets, with experimental results demonstrating the effectiveness and robustness of CityScapeEye.

O treinamento de sistemas de Tradução de Máquina baseada em Estatística (TME) é bastante dependente da disponibilidade de corpora paralelos. Entretanto, este tipo de recurso costuma ser difícil de ser encontrado, especialmente quando lida com idiomas com poucos recursos ou com tópicos muito específicos, como, por exemplo, dermatologia. Para contornar esta situação, uma possibilidade é utilizar corpora comparáveis, que são recursos muito mais abundantes. Um modo de adquirir corpora comparáveis é a aplicação de algoritmos de Coleta Focada (CF). Neste trabalho, são propostas novas abordagens para CF, algumas baseadas em n-gramas e outras no poder expressivo das expressões multipalavra. Também são avaliadas a viabilidade do uso de CF para realização de adaptação de domínio para sistemas genéricos de TME e se há alguma correlação entre a qualidade dos algoritmos de CF e dos sistemas de TME que podem ser construídos a partir dos respectivos dados coletados. Os resultados indicam que algoritmos de CF podem ser bons meios para adquirir corpora comparáveis para realizar adaptação de domínio para TME e que há uma correlação entre a qualidade dos dois processos. / Statistical Machine Translation (SMT) is highly dependent on the availability of parallel corpora for training. However, these kinds of resource may be hard to be found, especially when dealing with under-resourced languages or very specific domains, like the dermatology. For working this situation around, one possibility is the use of comparable corpora, which are much more abundant resources. One way of acquiring comparable corpora is to apply Focused Crawling (FC) algorithms. In this work we propose novel approach for FC algorithms, some based on n-grams and other on the expressive power of multiword expressions. We also assess the viability of using FC for performing domain adaptations for generic SMT systems and whether there is a correlation between the quality of the FC algorithms and of the SMT systems that can be built with its collected data. Results indicate that the use of FCs is, indeed, a good way for acquiring comparable corpora for SMT domain adaptation and that there is a correlation between the qualities of both processes.

Sentiment analysis concerns the computational study of opinions expressed in text. Social media domains provide a wealth of opinionated data, thus, creating a greater need for sentiment analysis. Typically, sentiment lexicons that capture term-sentiment association knowledge are commonly used to develop sentiment analysis systems. However, the nature of social media content calls for analysis methods and knowledge sources that are better able to adapt to changing vocabulary. Invariably existing sentiment lexicon knowledge cannot usefully handle social media vocabulary which is typically informal and changeable yet rich in sentiment. This, in turn, has implications on the analyser's ability to effectively capture the context therein and to interpret the sentiment polarity from the lexicons. In this thesis we use SentiWordNet, a popular sentiment-rich lexicon with a substantial vocabulary coverage and explore how to adapt it for social media sentiment analysis. Firstly, the thesis identifies a set of strategies to incorporate the effect of modifiers on sentiment-bearing terms (local context). These modifiers include: contextual valence shifters, non-lexical sentiment modifiers typical in social media and discourse structures. Secondly, the thesis introduces an approach in which a domain-specific lexicon is generated using a distant supervision method and integrated with a general-purpose lexicon, using a weighted strategy, to form a hybrid (domain-adapted) lexicon. This has the dual purpose of enriching term coverage of the general purpose lexicon with non-standard but sentiment-rich terms as well as adjusting sentiment semantics of terms. Here, we identified two term-sentiment association metrics based on Term Frequency and Inverse Document Frequency that are able to outperform the state-of-the-art Point-wise Mutual Information on social media data. As distant supervision may not be readily applicable on some social media domains, we explore the cross-domain transferability of a hybrid lexicon. Thirdly, we introduce an approach for improving distant-supervised sentiment classification with knowledge from local context analysis, domain-adapted (hybrid) and emotion lexicons. Finally, we conduct a comprehensive evaluation of all identified approaches using six sentiment-rich social media datasets.

Cette thèse s’intéresse à l’adaptation d’un classifieur statistique d’opinion au niveau du texte d’un domaine à un autre. Cependant, nous exprimons notre opinion différemment selon ce dont nous parlons. Un même mot peut ne pas désigner pas la même chose ou bien ne pas avoir la même connotation selon le thème de la discussion. Si ces mots ne sont pas détectés, ils induiront des erreurs de classification.Nous appelons donc marqueurs multi-polaires des mots ou bigrammes dont la présence indique une certaine polarité du texte entier, différente selon le domaine du texte. Cette thèse est consacrées à leur étude. Ces marqueurs sont détectés à l’aide d’un test du khi2 lorsque l’on dispose d’annotations au niveau du texte dans les deux domaines d’intérêt. Nous avons également proposé une méthode de détection semi-supervisé. Nous utilisons une collections de mots pivots auto-épurés afin d’assurer une polarité stable d’un domaine à un autre.Nous avons également vérifié la pertinence linguistique des mots sélectionnés en organisant une campagne d’annotation manuelle. Les mots ainsi validés comme multi-polaires peuvent être des éléments de contexte, des mots exprimant ou expliquant une opinion ou bien désignant l’objet sur lequel l’opinion est portée. Notre étude en contexte a également mis en lumière trois causes principale de changement de polarité : le changement de sens, le changement d’objet et le changement d’utilisation.Pour finir, nous avons étudié l’influence de la détection des marqueurs multi-polaires sur la classification de l’opinion au niveau du texte par des classifieurs automatiques dans trois cas distincts : adaptation d’un domaine source à un domaine cible, corpus multi-domaine, corpus en domaine ouvert. Les résultats de ces expériences montrent que plus le transfert initial est difficile, plus la prise en compte des marqueurs multi-polaires peut améliorer la classification, allant jusqu’à plus cinq points d’exactitude. / In this thesis, we are studying the adaptation of a text level opinion classifier across domains. Howerver, people express their opinion in a different way depending on the subject of the conversation. The same word in two different domains can refer to different objects or have an other connotation. If these words are not detected, they will lead to classification errors.We call these words or bigrams « multi-polarity marquers ». Their presence in a text signals a polarity wich is different according to the domain of the text. Their study is the subject of this thesis. These marquers are detected using a khi2 test if labels exist in both targeted domains. We also propose a semi-supervised detection method for the case with labels in only one domain. We use a collection of auto-epurated pivot words in order to assure a stable polarity accross domains.We have also checked the linguistic interest of the selected words with a manual evaluation campaign. The validated words can be : a word of context, a word giving an opinion, a word explaining an opinion or a word wich refer to the evaluated object. Our study also show that the causes of the changing polarity are of three kinds : changing meaning, changing object or changing use.Finally, we have studyed the influence of multi-polarity marquers on opinion classification at text level in three different cases : adaptation of a source domain to a target domain, multi-domain corpora and open domain corpora. The results of our experiments show that the potential improvement is bigger when the initial transfer was difficult. In the favorable cases, we improve accurracy up to five points.

abstract: Humans have a great ability to recognize objects in different environments irrespective of their variations. However, the same does not apply to machine learning models which are unable to generalize to images of objects from different domains. The generalization of these models to new data is constrained by the domain gap. Many factors such as image background, image resolution, color, camera perspective and variations in the objects are responsible for the domain gap between the training data (source domain) and testing data (target domain). Domain adaptation algorithms aim to overcome the domain gap between the source and target domains and learn robust models that can perform well across both the domains. This thesis provides solutions for the standard problem of unsupervised domain adaptation (UDA) and the more generic problem of generalized domain adaptation (GDA). The contributions of this thesis are as follows. (1) Certain and Consistent Domain Adaptation model for closed-set unsupervised domain adaptation by aligning the features of the source and target domain using deep neural networks. (2) A multi-adversarial deep learning model for generalized domain adaptation. (3) A gating model that detects out-of-distribution samples for generalized domain adaptation. The models were tested across multiple computer vision datasets for domain adaptation. The dissertation concludes with a discussion on the proposed approaches and future directions for research in closed set and generalized domain adaptation. / Dissertation/Thesis / Masters Thesis Computer Science 2020

Fault detection and diagnostics are crucial tasks in condition-based maintenance. Industries nowadays are in need of fault identification in their machines as early as possible to save money and take precautionary measures in case of fault occurrence. Also, it is beneficial for the smooth interference in the manufacturing process in which it avoids sudden malfunctioning. Having sufficient training data for industrial machines is also a major challenge which is a prerequisite for deep neural networks to train an accurate prediction model. Transfer learning in such cases is beneficial as it can be helpful in adapting different operating conditions and characteristics which is the casein real-life applications. Our work is focused on a pneumatic system which utilizes compressed air to perform operations and is used in different types of machines in the industrial field. Our novel contribution is to build upon a Domain Adversarial Neural Network (DANN) with a unique approach by incorporating ensembling techniques for diagnostics of air leakage problem in the pneumatic system under transfer learning settings. Our approach of using ensemble methods for feature extraction shows up to 5 % improvement in the performance. We have also performed a comparative analysis of our work with conventional machine and deep learning methods which depicts the importance of transfer learning and we have also demonstrated the generalization ability of our model. Lastly, we also mentioned a problem specific contribution by suggesting a feature engineering approach, such that it could be implemented on almost every pneumatic system and could potentially impact the prediction result positively. We demonstrate that our designed model with domain adaptation ability will be quite useful and beneficial for the industry by saving their time and money and providing promising results for this air leakage problem in the pneumatic system.

With the increasing demand for labeled data in machine learning for visual perception tasks, the interest in using synthetically generated data has grown. Due to the existence of a domain gap between synthetic and real data, strategies in domain adaptation are necessary to achieve high performance with models trained on synthetic or mixed data. With a dataset of synthetically blocked fish-eye lenses in traffic environments, we explore different strategies to train a neural network. The neural network is a binary classifier for full blockage detection. The different strategies tested are data mixing, fine-tuning, domain adversarial training, and adversarial discriminative domain adaptation. Different ratios between synthetically generated data and real data are also tested. Our experiments showed that fine-tuning had slightly superior results in this test environment. To fully take advantage of the domain adversarial training, training until domain indiscriminate features are learned is necessary and helps the model attain higher performance than using random data mixing.

abstract: In recent years, the rise in social media usage both vertically in terms of the number of users by platform and horizontally in terms of the number of platforms per user has led to data explosion. User-generated social media content provides an excellent opportunity to mine data of interest and to build resourceful applications. The rise in the number of healthcare-related social media platforms and the volume of healthcare knowledge available online in the last decade has resulted in increased social media usage for personal healthcare. In the United States, nearly ninety percent of adults, in the age group 50-75, have used social media to seek and share health information. Motivated by the growth of social media usage, this thesis focuses on healthcare-related applications, study various challenges posed by social media data, and address them through novel and effective machine learning algorithms. The major challenges for effectively and efficiently mining social media data to build functional applications include: (1) Data reliability and acceptance: most social media data (especially in the context of healthcare-related social media) is not regulated and little has been studied on the benefits of healthcare-specific social media; (2) Data heterogeneity: social media data is generated by users with both demographic and geographic diversity; (3) Model transparency and trustworthiness: most existing machine learning models for addressing heterogeneity are considered as black box models, not many providing explanations for why they do what they do to trust them. In response to these challenges, three main research directions have been investigated in this thesis: (1) Analyzing social media influence on healthcare: to study the real world impact of social media as a source to offer or seek support for patients with chronic health conditions; (2) Learning from task heterogeneity: to propose various models and algorithms that are adaptable to new social media platforms and robust to dynamic social media data, specifically on modeling user behaviors, identifying similar actors across platforms, and adapting black box models to a specific learning scenario; (3) Explaining heterogeneous models: to interpret predictive models in the presence of task heterogeneity. In this thesis, novel algorithms with theoretical analysis from various aspects (e.g., time complexity, convergence properties) have been proposed. The effectiveness and efficiency of the proposed algorithms is demonstrated by comparison with state-of-the-art methods and relevant case studies. / Dissertation/Thesis / Doctoral Dissertation Computer Science 2019