The Experience Sampling Method (ESM) is increasingly recognized for its ability to capture fine-grained, real-time insights into individuals’ emotional and behavioral states in their everyday environments. While the utility of ESM in clinical contexts has been well-documented, its integration into practical tools for mental health professionals remains underexplored. This study investigates how meaningful insights can be derived and presented from ESM data to assist clinicians in the identification of mental health symptoms and intervention planning. A series of prototype visualizations, in the form of line charts, pie and bar charts and network diagrams, were developed and then evaluated for usefulness, user-friendliness and intuitiveness by a total of 8 participants. Results suggest that no single visualization is sufficient on its own, but that a combination of temporal, contextual, and relational visualizations provides a more complete view of a patient’s emotional patterns. While bar charts were preferred for clarity and comparison, pie charts were appreciated for quickly identifying dominant factors by a few participants. Network diagrams, though initially less intuitive, were still valued when paired with descriptive captions. These findings support the potential of multi-format visual tools to assist clinicians in extracting meaningful insights from ESM data.

Smoking cessation remains a persistent public health challenge, with digital interventions increasingly adopted to support behavior change. This study explores how smokers plan and engage with preparatory activities suggested by conversational agents in online cessation programs. It addresses three main questions: the factors that influence engagement with these activities, the ability of large language models to identify smokers’ articulated plans, and the role of conditional "if-then" formulations in expressing coping strategies. A thematic analysis was conducted on qualitative user responses using both manual coding and automated labeling by local large language models. The findings show that smokers create context-sensitive plans shaped by emotional states, routines, and perceived usefulness of the suggestions, often expressed through conditional intentions. While manual analysis produced consistent and detailed themes, the large language model showed low agreement, highlighting limitations in current AI driven qualitative analysis. These results inform the design of more adaptive digital cessation tools and contribute to the understanding of AI’s role in supporting thematic research.

The use of Artificial Social Agents (ASAs) is rapidly expanding across society. As these agents become more integrated into our interactions, understanding the user experience of them becomes increasingly necessary to ensure their design aligns with user needs, promotes trust, and supports meaningful engagement. This study aims to investigate how users experience interactions with ASAs, focusing on using thematic analysis to identify recurring themes in user-reported experiences with ASAs. In addition, it also explores the reliability of locally hosted Large Language Models (LLMs) in identifying those experiences. We conducted a manual -peer validated- thematic analysis, resulting in a total of 31 themes. Afterwards, we conducted two experiments with LLMs, namely giving hem an unguided prompt (i.e. the LLM discovers and groups themes independently) and a guided prompt (i.e. the LLM matches predefined themes to responses) and measured their agreements with the manual analysis both intuitively and analytically. From our findings, it became clear that users experience ASAs through a balance of practical utility and emotional engagement. Themes covering the agent's helpfulness, sociability, enjoyability and perceived intelligence played a central role in shaping user experience. Most users responded positively to ASAs that felt intuitive, responsive, and human-like, though perceptions of human-likeness varied, sometimes enhancing the experience and other times creating discomfort. Our evaluation of LLMs showed that while they are capable of uncovering broad thematic patterns through unguided analysis, they fall short when tasked with consistently identifying and labeling predefined themes at the individual response level. This suggests that current LLMs, while useful as supplementary tools, are not yet reliable replacements for human-led thematic analysis in capturing the full nuance of user experiences at a detailed level. The conclusions reinforce the continued value and need of human-led thematic analysis, particularly when aiming to capture subtle, context-dependent insights that automated models may overlook.

Digital smoking cessation tools increasingly use chatbots to recommend preparatory activities intended to support behavior change. However, little is known about how users actually experience these activities and whether large language models (LLMs) can support the qualitative analysis of such experiences. This study explores how smokers and vapers described their experiences with chatbot-suggested activities in an online intervention and investigates whether an LLM can assist with analyzing their responses.

We conducted a reflexive thematic analysis on 650 filtered user responses and examined whether a local version of LLaMA 3.3 (8B) could support the generation of common patterns and themes from qualitative data, and the categorization of such data into predefined themes. Our findings show that participants had mixed experiences. Some felt encouraged and supported by the activities, while others found them confusing, irrelevant to smoking cessation, or difficult to complete due to contextual or environmental barriers.

While the model was able to generate reasonable themes that overlapped with human interpretations, it was unable to reliably label data points using a predefined coding schema, achieving a Cohen’s Kappa of just 0.003. These findings suggest that LLMs may be useful in the early, exploratory stages of qualitative analysis, but currently lack the accuracy needed for theme application in complex, nuanced data.

Smoking cessation interventions sometimes involve the use of both technology and human support to increase effectiveness. Nevertheless, little is known about user preferences for allocating human support and whether large language models (LLMs) can support qualitative analysis to better understand these preferences. This study analyzes how smokers’ ethical perspectives shape their preferences for time allocation mechanisms in online smoking cessation programs, while also evaluating how can LLMs support this analysis.

We conducted a deductive thematic analysis of open-ended responses from users who completed a questionnaire after participating in a smoking cessation program with a virtual coach. Additionally, we employed the LLaMA large language model to identify patterns in the responses and to assign the ethical themes discovered during the analysis.

The findings indicate that some users valued fairness and preferred scheduled, randomized interventions or no feedback. Others emphasized autonomy, wanting users to request feedback themselves. Some suggested prioritizing motivated or advanced users. A common view was that interventions should focus on those in need, users at risk of disengagement or health problems, those making little progress, experiencing emotional difficulties, or lacking clarity. The large language model was successful in identifying themes but not in accurately allocating themes, reflected by a low Cohen’s Kappa of 0.05.

The results presenting user preferences can guide the design of interventions that need to be effective and ethically sound. The findings suggest that while large language models can identify themes, they are not yet suitable for allocating those themes.

Side-channel attacks (SCA) play a crucial role in assessing the security of the implementation of cryp- tographic algorithms. Still, traditional profiled attacks require a nearly identical reference device to the target, limiting their practicality. This thesis focuses on non-profiled SCA, which provides a re- alistic alternative when the attacker lacks access to a profiling device. Specifically, we investigate non-profiled deep learning-based SCA techniques. Our evaluation first explores existing unsupervised deep learning-based side-channel analysis approaches: differential deep learning analysis (DDLA) and multi-output regression (MOR). We show that using a validation set for key distinguishing, rather than the training set, improves the overall success rate across various datasets.
In the context of multi-output regression SCA, we comprehensively evaluate different loss functions. The thesis proposes a novel approach that outperforms existing methods by employing the normalized Z-score MSE (Z-MSE) loss function. Additionally, we introduce two key distinguishing methods, one based on the smallest Z-MSE loss and the other on the highest Pearson correlation between actual and predicted labels during validation. The experimental results show the efficacy of the novel approach in breaking traces protected by countermeasures. Notably, even with high levels of desynchronization (250) for ASCADf traces, the attack succeeds within a limited number of epochs (15).
Moreover, we demonstrate that the performance of the novel approach can be further enhanced through ensembles. This leads to a reduction in the number of traces required to break the key for most datasets and a decrease in the average guessing entropy. Data augmentation also proves beneficial in some instances, resulting in improved success rates.

Side-Channel Attacks (SCA) attempt to recover the secret cryptographic key from an electronic device by exploiting the unintended physical leakages of said device. With the devices that are being attacked becoming more sophisticated, so is SCA. In the past few years, the focus of the research in the field of SCA shifted towards the application of the powerful method of Deep Learning (DL). DL SCA methods can operate in a similar way as before seen methods in the profiled setting, such as Template Attack. In the profiled setting, SCA first creates a profile on a copy of the target device, to then subsequently use that same profile to perform a more powerful attack on the target device. DL SCA has proven to be quite the effective method, even showcasing its success against implementations utilising countermeasures. However, as DL SCA is fairly novel there still exist gaps in the knowledge we have of how to make DL SCA effective in all possible situations. Most research bases itself on software-based implementations, whilst rarely hardware-based implementations are discussed as they are often seen as more difficult to attack. Our contribution in this work is to showcase the difference in difficulty between hardware-based implementations and software-based implementations that both use countermeasures. We explore the attack performance of several state-of-the-art methods on hardware-based implementations with countermeasures and give insight into why their performance is the way it is. We also attempt to make a base methodology for attacking hardware-based implementations, both with and without countermeasures, as the current field of research is lacking this. Showcasing our suggested methodology, we achieve better than state-of-the-art results on a hardware-based implementation with countermeasures and competitive with the state-of-the-art results on a hardware-based implementation without countermeasures.

Some of the most prominent types of attacks against modern cryptographic implementations are side-channel attacks. These attacks leverage some unintended, often physical, leakage of the implementation to retrieve secret information. In recent times, a large part of the focus of side-channel research has been on deep learning methods. These methods operate in a profiled setting where a model is learned based on a copy of the device that is being attacked. This model is subsequently used to create significantly more potent attacks against the target. Attacks using deep learning methods can often defeat even implementations protected with countermeasures, but as implementations become more protected, novel methods are required to successfully generate attacks.
Recently, residual neural networks have been used for side-channel attacks, and these networks show promising attacking performance. However, these novel networks are relatively limited, and a more thorough investigation into the construction of residual networks in the side-channel context is required.
Our contribution is a more thorough investigation into the construction of these residual architectures. We explore several important factors to the construction of these models and generate insights into various methods for this construction. The resulting architectures we find show attacking performance that is competitive with the state-of-the-art methods across various data sets and feature selection scenarios.

Side-channel attacks (SCA) can obtain secret information related to the private key used during encryption executed on some device by exploiting leakage in power traces produced by the device. In recent years, researchers found that a neural network (NN) can be employed to execute a powerful profiled SCA, even on targets protected with countermeasures. This paper explores the effectiveness of (1) utilising a genetic algorithm (GA) to train the weights of NNs used for SCA, (2) using NeuroEvolution of Augmenting Topologies (NEAT), which is a commonly used approach for automated NN architecture search, to construct architectures for use in SCA, and (3) \textit{Neuroevolution to Attack Side-Channel Traces Yielding Convolutional Neural Networks} (NASCTY-CNNs), a novel GA that applies genetic operators on the scale of layers rather than neurons to automatically produce CNNs for side-channel analysis. The results indicate that weight evolution and our neat-based approach are successful when attacking weak, unprotected targets. However, they do not perform well on the ASCAD data set, against which SCA methods are commonly benchmarked, suggesting these methods require significant tweaking to be effective. NASCTY can automatically produce novel NNs that achieve performance close to state-of-the-art approaches on both masked and desynchronised ASCAD traces, demonstrating that such neuroevolution methods provide a solid venue for further research.

Side-channel attacks (SCA) focus on vulnerabilities caused by insecure implementations and exploit them to deduce useful information about the data being processed or the data itself through leakages obtained from the device. There have been many studies exploiting these side-channel leakages, and most of the state-of-the-art attacks have been shown to work on systems implementing AES. The methodology is usually based on exploiting leakages for the outer rounds, i.e., the first and the last round. In some cases, due to partial countermeasures or the nature of the device itself, it might not be possible to attack the outer round leakages. In this case, the attacker has to resort to attacking the inner rounds. This work provides a generalization for inner round side-channel attacks on AES and PRESENT, and experimentally validates the same for AES with non-profiled and profiled attacks. We formulate the computation of the hypothesis values of any byte in the intermediate rounds of both AES and PRESENT. The more inner the round is, the higher is the attack complexity in terms of the number of bits to be guessed for the hypothesis. We discuss the main limitations for obtaining predictions in inner rounds and, in particular, we compare the performance of Correlation Power Analysis (CPA) against deep learning-based profiled side-channel attacks (DL-SCA). We demonstrate that because trained deep learning models require fewer traces in the attack phase, they also have fewer complexity limitations to attack inner AES rounds than non-profiled attacks such as CPA.

Clinical decision support systems is a collection name for a lot of Artificial Intelligence systems used in healthcare. These systems are designed to help health workers make decisions faster and make the healthcare environment as a whole more efficient. Decisions made by these systems often weigh heavy on the ethical side giving advice on what kind of care a patient receives. The ethics of these decision mean it is even more important for these systems to act fair and unbiased towards all patients. This however is not always the case though. This paper will explain and dissect the issues of unfairness within clinical decision support systems and will compare and adapt different socio-technical design methodologies to form a advice on designing for fair clinical decision support algorithms.

Companies have been looking for automation in their hiring practices and Artificial Intelligence offers a solution. A popular opinion on the topic is that AI-enabled hiring will enhance talent acquisition overall, eliminating bias and bettering results. However, deployed AI hiring tools often do not offer a solution but worsen the problem by violating ethical and moral barriers. Thus, it is crucial to consider the socio-technical perspective of this topic when designing software systems integrated into hiring practices. As part of AI-enabled recruitment, CV screening software is deployed in numerous conglomerates worldwide and is the centre of attention of this paper. The theoretical results delivered via this research originate from a variety of scientific articles of a multi-disciplinary nature. This paper examines how CV screening software can be designed to estimate professional proficiency while taking into consideration the needs and moral values of the stakeholders involved. A literature study has been conducted to derive the results. The research examines three design methodologies (Systemic design, Value Sensitive Design and Human-centered design) that help the design team to address the aforementioned issues and considers their impact on the design process of a CV screening tool regarding proficiency evaluation. The results of the paper conclude that it is crucial to assemble a design team that is multi-disciplinary in nature to elicit and embed the stakeholders' values into the design of a technical system of this nature.

Artificial Intelligence (AI) is widely used in hiring practices to identify the most suitable candidate for a vacancy. This is due to the promise of higher overall efficiency and lower costs. However, these AI-powered tools may create an inaccurate conception of the applicant's suitability to the vacancy by numerically quantifying context-dependent variables. If only this inaccurate conception is used to judge an applicant, this is a violation of the applicant's autonomy over their self-representation. This paper argues that such a problem could be solved by adopting a broader design scope - Socio-Technical Systems Design (STSD). STSD approaches have not been widely applied to AI or hiring practices yet. Therefore the main contribution of this paper is to bridge this gap by exploring possible STSD approaches which can be applied to ensure the applicants' autonomy over self-representation. This paper suggests combining methodologies and design principles from two STSD approaches - Design for Values and Systemic Design. The findings from Design for Values suggest that key stakeholders should be involved in the design process. Therefore, the designers should conduct a stakeholder analysis to identify the key stakeholders, followed by an investigation to explore the stakeholders' needs, and the values which the proposed system could implicate. Systemic Design offers design principles that should be utilized during the design process. These principles consist of: expanding the problem space; focus on the relationship between the key stakeholders; and follow an iterative, experimental, and evolutionary design approach throughout the design process. This nuanced, stakeholder-centric approach results in an inclusive, transparent, multi-discipline, and socially aware process which is necessary to understand the complex social context, and its socio-ethical issues.

Machine learning algorithms were used in in the past decade to assist humans with recruitment and grades assessments in the academic field. For the most part, the algorithms either exacerbated existing biases or output unfair results. This could often be traced back to an ill-implementation of the systems in the social context. The academic admission process is defined as setting goals, locating candidates, ranking and accepting them. To properly integrate machine learning in such process, one may follow the Value Sensitive methodology, which suggests designing a technical system around a social value. This methodology takes into account the various stakeholders, values and technical solutions available. Later, the system should be iteratively improved and constantly evaluated and examined so that it still serves the core values as defined.

The cyber arms race has red and blue teams continuously at their toes to keep ahead. Increasingly capable cyber actors breach secure networks at a worrying scale. While network monitoring and analysis should identify blatant data exfiltration attempts, covert channels bypass these measures and facilitate surreptitious information extraction. The many legitimate uses and widespread availability of DNS, the "phone book" of the internet, make it an attractive protocol for such covert channels. Covert DNS storage channels encode information in the payload of outbound DNS queries.

This thesis aims to assess the effectiveness of using machine learning methods to detect covert DNS storage channels. Our literature survey identified distinct differences in 1) algorithm type, either unsupervised anomaly detection or supervised classification, and 2) the information source for features, either isolated DNS queries or query sequences.

We performed experiments with (Extended) Isolation Forest algorithms for anomaly detection and Random Forests for classification, combined with different feature set compositions to evaluate their relative performance. Payload-only features were derived from isolated queries and behavioral features were extracted from time-based or fixed-length sliding windows over per-domain query sequences. We evaluated our models using a large-scale corporate DNS dataset of real-world proportions and a novel dataset of connection tunneling traffic and simulated credit card exfiltration malware.

We found that the majority of experiments were able to achieve high detection rates of 98.6% or more on a variety of storage channel threats, at low false positive rates. Classification models significantly outperform anomaly detection models on threats seen during training. Evaluation on unseen threats, however, revealed that generalization is difficult, provided the limited set of training threats and showed anomaly detection models more capable at detecting a variety of threats than classification models. We furthermore showed that feature sets with a behavioral component consistently outperform payload­-only features, although our experiments were inconclusive regarding the relative performance between composite feature sets.

Given the prevalence of benign storage channels misusing DNS for legitimate data transfer, we recommend rigorous filtering of training data beforehand to improve model optimization and evaluation. Furthermore, extending the malicious training set with DNS command-and-control (C2) malware is a promising future research direction to improve generalization of classification models.

Over the last decades, side-channel attacks (SCAs) have been proven as a substantial weakness of cryptographic devices, while the recent growth of deep learning (DL) dramatically improved the performance of SCA. More and more researches present ways to build lightweight deep neural network (DNN) models that can retrieve the secret encryption key by analyzing a few power traces of the captured devices. In the meantime, traditional countermeasures are rendered more or less ineffective against these sophisticated SCA.

This research aims to present a novel approach in building SCA countermeasures using adversarial examples, cleverly crafted inputs that trick DNN models and force them to misclassify. As modern SCAs rely on DNN models, an adversarial-based countermeasure could alter the side-channel leakage so that the attacking model cannot identify the correct key. To investigate this approach, we add artificial adversarial noise on an unprotected dataset using evolutionary computation techniques and see how SCAs are affected. We present different methods on how this noise can be designed and how the different parameters of each method affect its performance. Our results indicate that an adversarial-based countermeasure can decrease sufficiently the performance of the attack.

Furthermore, we compare our adversarial-based countermeasures with traditional countermeasures. We show that our proposal can provide equivalent or better protection in some cases, while it presents worse performance than others. Through our systematic research, we identify the reasons behind this weakness and find solutions for future work.

Additionally, we investigate how such a countermeasure can perform against non-profiled SCAs. We show that our adversarial-based countermeasure is effective against this specific family of SCAs, even though it is designed against DNN-based attacks. Comparing this performance to traditional SCA countermeasures, we see that our proposal can be an alternative to noise-based countermeasures, while it fails to provide better security compared to time-based countermeasures.