Automated vehicles (AVs) are promising applications of artificial intelligence (AI). While human drivers benefit from AVs, including long-distance support and collision prevention, we do not always understand how AV systems function and make decisions. Consequently, drivers might develop inaccurate mental models and form unrealistic expectations of these systems, leading to unwanted incidents. Although efforts have been made to support drivers' understanding of AVs through in-vehicle visual and auditory interfaces and warnings, these may not be sufficient or effective in addressing user confusion and overtrust in in-vehicle technologies, sometimes even creating negative experiences. To address this challenge, this dissertation conducts a series of studies to explore the possibility of using the in-vehicle intelligent agent (IVIA) in the form of the speech user interface to support drivers, aiming to enhance safety, performance, and satisfaction in conditionally automated vehicles. First, two expert workshops were conducted to identify design considerations for general IVIAs in the driving context. Next, to better understand the effectiveness of different IVIA designs in conditionally automated driving, a driving simulator study (n=24) was conducted to evaluate four types of IVIA designs varying by embodiment conditions and speech styles. The findings indicated that conversational agents were preferred and yielded better driving performance, while robot agents caused greater visual distraction. Then, contextual inquiries with 10 drivers owning vehicles with advanced driver assistance systems (ADAS) were conducted to identify user needs and the learning process when interacting with in-vehicle technologies, focusing on interface feedback and warnings. Subsequently, through expert interviews with seven experts from AI, social science, and human-computer interaction domains, design considerations were synthesized for improving the explainability of AVs and preventing associated risks. With information gathered from the first four studies, three types of adaptive IVIAs were developed based on human-automation function allocation and investigated in terms of their effectiveness on drivers' response time, driving performance, and subjective evaluations through a driving simulator study (n=39). The findings indicated that although drivers preferred more information provided to them, their response time to road hazards might be degraded when receiving more information, indicating the importance of the balance between safety and satisfaction. Taken together, this dissertation indicates the potential of adopting IVIAs to enhance the explainability of future AVs. It also provides key design guidelines for developing IVIAs and constructing explanations critical for safer and more satisfying AVs. / Doctor of Philosophy / Automated vehicles (AVs) are an exciting application of artificial intelligence (AI). While these vehicles offer benefits like helping with long-distance driving and preventing accidents, people often do not understand how they work or make decisions. This lack of understanding can lead to unrealistic expectations and potentially dangerous situations. Even though there are visual and sound alerts in these cars to help drivers, they are not always sufficient to prevent confusion and over-reliance on technology, sometimes making the driving experience worse. To address this challenge, this dissertation explores the use of in-vehicle intelligent agents (IVIAs), in the form of speech assistant, to help drivers better understand and interact with AVs, aiming to improve safety, performance, and overall satisfaction in semi-automated vehicles. First, two expert workshops helped identify key design features for IVIAs. Then, a driving simulator study with 24 participants tested four different designs of IVIAs varying in appearance and how they spoke. The results showed that people preferred conversational agents, which led to better driving behaviors, while robot-like agents caused more visual distractions. Then, through contextual inquiries with 10 drivers who own vehicles with advanced driver assistance systems (ADAS), I identified user needs and how they learn to interact with in-car technologies, focusing on feedback and warnings. Subsequently, I conducted expert interviews with seven professionals from AI, social science, and human-computer interaction fields, which provided further insights into facilitating the explainability of AVs and preventing associated risks. With the information gathered, three types of adaptive IVIAs were developed based on whether the driver was actively in control of the vehicle, or the driving automation system was in control. The effectiveness of these agents was evaluated through drivers' brake and steer response time, driving performance, and user satisfaction through another driving simulator study with 39 participants. The findings indicate that although drivers appreciated more detailed explanations, their response time to road hazards slowed down, highlighting the need to balance safety and satisfaction. Overall, this research shows the potential of using IVIAs to make AVs easier to understand and safer to use. It also offers important design guidelines for creating these IVIAs and their speech contents to improve the driving experience.

<p dir="ltr">The study builds and evaluates three advanced neural network models—ResNet-34, EfficientNet B0, and SqueezeNet—for defect detection and classification of silicon wafer map images. The study evaluates the neural network model in two cases, binary and multi-class classifications. The binary classification, which is crucial for promptly determining whether a wafer map is defective, EfficientNet-B0 led with the highest test accuracy of 94.62% and an average accuracy of 93.2%. Similarly, in multi-class classification, necessary for pinpointing specific defect causes early in the manufacturing process, EfficientNet-B0 achieved the top test accuracy of 84.22% with an average accuracy of 84.07%. Further enhancements in the study resulted from strategic pruning of EfficientNet-B0, specifically the removal of Residual Block 2 after convolutional layer visualization revealed minimal impact on accuracy, with a reduction of just 1.33%. These modifications not only refined the learning process but also reduced the model size by 33%, thereby increasing computational efficiency. The integration of Grad-CAM++ visualizations ensured the model focused on pertinent features, thus boosting the transparency and reliability of the defect detection process. The results underscore the potential of advanced neural networks to significantly enhance the accuracy and efficiency of semiconductor manufacturing.</p>

The growing presence of algorithm-generated recommendations in AI-powered services highlights the importance of responsible systems that explain outputs in a human-understandable form, especially in an automotive context. Implementing explainability in recommendations of AI-powered eco-driving is important in ensuring that drivers understand the underlying reasoning behind the recommendations. Previous literature on explainable AI (XAI) has been primarily technological-centered, and only a few studies involve the end-user perspective. There is a lack of knowledge of drivers' needs and requirements for explainability in an AI-powered eco-driving context. This study addresses the attributes that make a “satisfactory” explanation, i,e., a satisfactory interface between humans and AI. This study uses scenario-based interviews to understand the explainability attributes that influence truck drivers' intention to use eco-driving recommendations. The study used thematic analysis to categorize seven attributes into context-dependent (Format, Completeness, Accuracy, Timeliness, Communication) and generic (Reliability, Feedback loop) categories. The study contributes context-dependent attributes along three design dimensions: Presentational, Content-related, and Temporal aspects of explainability. The findings of this study present an empirical foundation into end-users' explainability needs and provide valuable insights for UX and system designers in eliciting end-user requirements.

Knowledge bases are employed in a variety of applications, from natural language processing to semantic web search; alas, in practice, their usefulness is hurt by their incompleteness. To address this issue, several techniques aim at performing knowledge base completion, of which embedding models are efficient, attain state-of-the-art accuracy, and eliminate the need for feature engineering. However, embedding models predictions are notoriously hard to interpret. In this work, we propose model-agnostic methods that allow one to interpret embedding models by extracting weighted Horn rules from them. More specifically, we show how the so-called \"pedagogical techniques\", from the literature on neural networks, can be adapted to take into account the large-scale relational aspects of knowledge bases, and show experimentally their strengths and weaknesses. / Bases de conhecimento apresentam diversas aplicações, desde processamento de linguagem natural a pesquisa semântica da web; contudo, na prática, sua utilidade é prejudicada por não serem totalmente completas. Para solucionar esse problema, diversas técnicas focam em completar bases de conhecimento, das quais modelos de embedding são eficientes, atingem estado da arte em acurácia, e eliminam a necessidade de fazer-se engenharia de características dos dados de entrada. Entretanto, as predições dos modelos de embedding são notoriamente difíceis de serem interpretadas. Neste trabalho, propomos métodos agnósticos a modelo que permitem interpretar modelos de embedding através da extração de regras Horn ponderadas por pesos dos mesmos. Mais espeficicamente, mostramos como os chamados \"métodos pedagógicos\", da literatura de redes neurais, podem ser adaptados para lidar com os aspectos relacionais e de larga escala de bases de conhecimento, e mostramos experimentalmente seus pontos fortes e fracos.

Knowledge bases are employed in a variety of applications, from natural language processing to semantic web search; alas, in practice, their usefulness is hurt by their incompleteness. To address this issue, several techniques aim at performing knowledge base completion, of which embedding models are efficient, attain state-of-the-art accuracy, and eliminate the need for feature engineering. However, embedding models predictions are notoriously hard to interpret. In this work, we propose model-agnostic methods that allow one to interpret embedding models by extracting weighted Horn rules from them. More specifically, we show how the so-called \"pedagogical techniques\", from the literature on neural networks, can be adapted to take into account the large-scale relational aspects of knowledge bases, and show experimentally their strengths and weaknesses. / Bases de conhecimento apresentam diversas aplicações, desde processamento de linguagem natural a pesquisa semântica da web; contudo, na prática, sua utilidade é prejudicada por não serem totalmente completas. Para solucionar esse problema, diversas técnicas focam em completar bases de conhecimento, das quais modelos de embedding são eficientes, atingem estado da arte em acurácia, e eliminam a necessidade de fazer-se engenharia de características dos dados de entrada. Entretanto, as predições dos modelos de embedding são notoriamente difíceis de serem interpretadas. Neste trabalho, propomos métodos agnósticos a modelo que permitem interpretar modelos de embedding através da extração de regras Horn ponderadas por pesos dos mesmos. Mais espeficicamente, mostramos como os chamados \"métodos pedagógicos\", da literatura de redes neurais, podem ser adaptados para lidar com os aspectos relacionais e de larga escala de bases de conhecimento, e mostramos experimentalmente seus pontos fortes e fracos.

An in-vehicle intrusion detection system (IV-IDS) is one of the protection mechanisms used to detect cyber attacks on electric or autonomous vehicles where anomaly-based IDS solution have better potential at detecting the attacks especially zero-day attacks. Generally, the IV-IDS generate false alarms (falsely detecting normal data as attacks) because of the difficulty to differentiate between normal and attack data. It can lead to undesirable situations, such as increased laxness towards the system, or uncertainties in the event-handling following a generated alarm. With the help of sophisticated Artificial Intelligence (AI) models, the IDS improves the chances of detecting attacks. However, the use of such a model comes at the cost of decreased interpretability, a trait that is argued to be of importance when ascertaining various other valuable desiderata, such as a model’s trust, causality, and robustness. Because of the lack of interpretability in sophisticated AI-based IV-IDSs, it is difficult for humans to trust such systems, let alone know what actions to take when an IDS flags an attack. By using tools found in the area of eXplainable AI (XAI), this thesis aims to explore what kind of explanations could be produced in accord with model predictions, to further increase the trustworthiness of AI-based IV-IDSs. Through a comparative survey, aspects related to trustworthiness and explainability are evaluated on a custom, pseudo-global, visualization-based explanation (”VisExp”), and a rule based explanation. The results show that VisExp increase the trustworthiness,and enhanced the  explainability of the AI-based IV-IDS.

Deep learning methods have the ability to accurately predict and interpret what data represents. However, the decision making of a deep learning model is not comprehensible for humans. This is a problem for sectors like market surveillance which needs clarity in the decision making of the used algorithms. This thesis aimed to investigate how a deep learning model can be constructed to make the decision making of the model humanly comprehensible, and to investigate the potential impact on classification performance. A literature study was performed and publicly available explanation methods were collected. The explanation methods LIME, SHAP, model distillation and SHAP TreeExplainer were implemented and evaluated on a ResNet trained on three different time-series datasets. A decision tree was used as the student model for model distillation, where it was trained with both soft and hard labels. A survey was conducted to evaluate if the explanation method could increase comprehensibility. The results were that all methods could improve comprehensibility for people with experience in machine learning. However, none of the methods could provide full comprehensibility and clarity of the decision making. The model distillation reduced the performance compared to the ResNet model and did not improve the performance of the student model. / Djupinlärningsmetoder har egenskapen att förutspå och tolka betydelsen av data. Däremot så är djupinlärningsmetoders beslut inte förståeliga för människor. Det är ett problem för sektorer som marknadsövervakning som behöver klarhet i beslutsprocessen för använda algoritmer. Målet för den här uppsatsen är att undersöka hur en djupinlärningsmodell kan bli konstruerad för att göra den begriplig för en människa, och att undersöka eventuella påverkan av klassificeringsprestandan. En litteraturstudie genomfördes och publikt tillgängliga förklaringsmetoder samlades. Förklaringsmetoderna LIME, SHAP, modelldestillering och SHAP TreeExplainer blev implementerade och utvärderade med en ResNet modell tränad med tre olika dataset. Ett beslutsträd användes som studentmodell för modelldestillering och den blev tränad på båda mjuka och hårda etiketter. En undersökning genomfördes för att utvärdera om förklaringsmodellerna kan förbättra förståelsen av modellens beslut. Resultatet var att alla metoder kan förbättra förståelsen för personer med förkunskaper inom maskininlärning. Däremot så kunde ingen av metoderna ge full förståelse och insyn på hur beslutsprocessen fungerade. Modelldestilleringen minskade prestandan jämfört med ResNet modellen och förbättrade inte prestandan för studentmodellen.

The thesis investigates how one should determine the appropriate transparency of an information processing system from a receiver perspective. Research in the past has suggested that the model should be maximally transparent for what is labeled as ”high stake decisions”. Instead of motivating the choice of a model’s transparency on the non-rigorous criterion that the model contributes to a high stake decision, this thesis explores an alternative method. The suggested method involves that one should let the transparency depend on how well an explanation of the model’s output satisfies the purpose of an explanation. As a result, we do not have to bother if it is a high stake decision, we should instead make sure the model is sufficiently transparent to provide an explanation that satisfies the expressed purpose of an explanation.

In the past few years, Artificial Intelligence (AI) has evolved into a powerful tool applied in multi-disciplinary fields to resolve sophisticated problems. As AI becomes more powerful and ubiquitous, oftentimes the AI methods also become opaque, which might lead to trust issues for the users of the AI systems as well as fail to meet the legal requirements of AI transparency. In this report, the possibility of making a credit-card fraud detection support system explainable to users is investigated through a quantitative survey. A publicly available credit card dataset was used. Deep Learning and Random Forest were the two Machine Learning (ML) methodsimplemented and applied on the credit card fraud dataset, and the performance of their results was evaluated in terms of their accuracy, recall, sufficiency, and F1 score. After that, two explainable AI (XAI) methods - SHAP (Shapley Additive Explanations) and LIME (Local Interpretable Model-Agnostic Explanations) were implemented and applied to the results obtained from these two ML methods. Finally, the XAI results were evaluated through a quantitative survey. The results from the survey revealed that the XAI explanations can slightly increase the users' impression of the system's ability to reason and LIME had a slight advantage over SHAP in terms of explainability. Further investigation of visualizing data pre-processing and the training process is suggested to offer deep explanations for users.

Online supervised learning from fast-evolving imbalanced data streams has applications in many areas. That is, the development of techniques that are able to handle highly skewed class distributions (or 'class imbalance') is an important area of research in domains such as manufacturing, the environment, and health. Solutions should be able to analyze large repositories in near real-time and provide accurate models to describe rare classes that may appear infrequently or in bursts while continuously accommodating new instances. Although numerous online learning methods have been proposed to handle binary class imbalance, solutions suitable for multi-class streams with varying degrees of imbalance in evolving streams have received limited attention. To address this knowledge gap, the first contribution of this thesis introduces the Online Learning from Imbalanced Multi-Class Streams through Dynamic Sampling (DynaQ) algorithm for learning in such multi-class imbalanced settings. Our approach utilizes a queue-based learning method that dynamically creates an instance queue for each class. The number of instances is balanced by maintaining a queue threshold and removing older samples during training. In addition, new and rare classes are dynamically added to the training process as they appear. Our experimental results confirm a noticeable improvement in minority-class detection and classification performance. A comparative evaluation shows that the DynaQ algorithm outperforms the state-of-the-art approaches. Our second contribution in this thesis focuses on fairness-aware learning from imbalanced streams. Our work is motivated by the observation that the decisions made by online learning algorithms may negatively impact individuals or communities. Indeed, the development of approaches to handle these concerns is an active area of research in the machine learning community. However, most existing methods process the data in offline settings and are not directly suitable for online learning from evolving data streams. Further, these techniques fail to take the effects of class imbalance, on fairness-aware supervised learning into account. In addition, recent fairness-aware online learning supervised learning approaches focus on one sensitive attribute only, which may lead to subgroup discrimination. In a fair classification, the equality of fairness metrics across multiple overlapping groups must be considered simultaneously. In our second contribution, we thus address the combined problem of fairness-aware online learning from imbalanced evolving streams, while considering multiple sensitive attributes. To this end, we introduce the Multi-Sensitive Queue-based Online Fair Learning (MQ-OFL) algorithm, an online fairness-aware approach, which maintains valid and fair models over evolving streams. MQ-OFL changes the training distribution in an online fashion based on both stream imbalance and discriminatory behavior of the model evaluated over the historical stream. We compare our MQ-OFL method with state-of-art studies on real-world datasets and present comparative insights on the performance. Our final contribution focuses on explainability and interpretability in fairness-aware online learning. This research is guided by the concerns raised due to the black-box nature of models, concealing internal logic from users. This lack of transparency poses practical and ethical challenges, particularly when these algorithms make decisions in finance, healthcare, and marketing domains. These systems may introduce biases and prejudices during the learning phase by utilizing complex machine learning algorithms and sensitive data. Consequently, decision models trained on such data may make unfair decisions and it is important to realize such issues before deploying the models. To address this issue, we introduce techniques for interpreting the outcomes of fairness-aware online learning. Through a case study predicting income based on features such as ethnicity, biological sex, age, and education level, we demonstrate how our fairness-aware learning process (MQ-OFL) maintains a balance between accuracy and discrimination trade-off using global and local surrogate models.