Abstract
Gastric cancer, ranking as the sixth most prevalent cancer globally and a leading cause of cancer-related mortality, follows a sequential progression known as Correa's cascade, spanning from chronic gastritis to eventual malignancy. Although endoscopy exams using NarrowBand Imaging are recommended by internationally accepted guidelines for diagnostic Gastric Intestinal Metaplasia, the lack of endoscopists with the skill to assess the NBI image patterns and the disagreement between endoscopists when assessing the same image, have made the use of biopsies the gold standard still used today. This proposal doctoral thesis seeks to address the challenge of developing a Computer-Aided Diagnosis solution for GIM detection in NBI endoscopy exams, aligning with the established guidelines, the Management of Epithelial Precancerous Conditions and Lesions in the Stomach. Our approach will involve a dataset creation that follows the standardized approach for histopathological classification of gastrointestinal biopsies, the Sydney System recommended by MAPS II guidelines, and annotation by gastroenterology experts. Deep learning models, including Convolutional Neural Networks, will be trained and evaluated, aiming to establish an internationally accepted AI-driven alternative to biopsies for GIM detection, promising expedited diagnosis, and cost reduction.

Abstract
Road safety is a global issue, with road-related accidents being one of the biggest leading causes of death. Motorcyclists are especially susceptible to injuries and death when there is an accident, due to the inherent characteristics of motorcycles. Accident prevention is paramount. To improve motorcycle safety, this paper discusses and proposes a preliminary architecture of a system composed of various sensors, to assist and warn the rider of potentially dangerous situations such as front and back collision warnings, pedestrian collision warnings, and road monitoring.

Abstract
In the field of gaming artificial intelligence, selecting the appropriate machine learning approach is essential for improving decision-making and automation. This paper examines the effectiveness of deep reinforcement learning (DRL) within interactive gaming environments, focusing on complex decision-making tasks. Utilizing the Unity engine, we conducted experiments to evaluate DRL methodologies in simulating realistic and adaptive agent behavior. A vehicle driving game is implemented, in which the goal is to reach a certain target within a small number of steps, while respecting the boundaries of the roads. Our study compares Proximal Policy Optimization (PPO) and Soft Actor-Critic (SAC) in terms of learning efficiency, decision-making accuracy, and adaptability. The results demonstrate that PPO successfully learns to reach the target, achieving higher and more stable cumulative rewards. Conversely, SAC struggles to reach the target, displaying significant variability and lower performance. These findings highlight the effectiveness of PPO in this context and indicate the need for further development, adaptation, and tuning of SAC. This research contributes to developing innovative approaches in how ML can improve how player agents adapt and react to their environments, thereby enhancing realism and dynamics in gaming experiences. Additionally, this work emphasizes the utility of using games to evolve such models, preparing them for real-world applications, namely in the field of vehicles' autonomous driving and optimal route calculation.

Abstract
This article presents a systematic review of the most prevalent vulnerabilities plaguing web and mobile applications. By analyzing recent research, it identifies a core set of vulnerabilities, including injection flaws, broken authentication, cross-site scripting (XSS), and insecure direct object references. Recognizing the human element, the article acknowledges the role of social engineering in exploiting these technical weaknesses. The review delves deeper, exploring how these vulnerabilities manifest differently across web and mobile platforms, considering factors like server-side security and API access. The research concludes by advocating for a defense strategy, emphasizing the importance of secure coding practices, robust authentication, and user awareness training. This comprehensive approach paves the way for a more secure digital landscape where both web and mobile applications can thrive. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.

Abstract
The Industry 4.0 paradigm (I4.0) supports the improvement of industrial processes through Information and Communication Technologies (ICT), with information systems providing real-time information to humans and machines, in order to make the production process more flexible and efficient. In this context, Virtual Assistants (VA) collect and process production data and provide contextualized and real-time information to the workers in the production environment. This paper presents a prototype of a VA developed to collect production data from heterogeneous sources in the factory, process them based on contextual information, and provide workers with useful information to assist them in taking informed decisions. In that context, VA can represent a valuable aid to improve overall productivity and efficiency in the I4.0 factories. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.

Abstract
Platforms to support novices learning to program are often accompanied by automated next-step hints that guide them towards correct solutions. Many of those approaches are data-driven, building on historical data to generate higher quality hints. Formal specifications are increasingly relevant in software engineering activities, but very little support exists to help novices while learning. Alloy is a formal specification language often used in courses on formal software development methods, and a platform-Alloy4Fun-has been proposed to support autonomous learning. While non-data-driven specification repair techniques have been proposed for Alloy that could be leveraged to generate next-step hints, no data-driven hint generation approach has been proposed so far. This paper presents the first data-driven hint generation technique for Alloy and its implementation as an extension to Alloy4Fun, being based on the data collected by that platform. This historical data is processed into graphs that capture past students' progress while solving specification challenges. Hint generation can be customized with policies that take into consideration diverse factors, such as the popularity of paths in those graphs successfully traversed by previous students. Our evaluation shows that the performance of this new technique is competitive with non-data-driven repair techniques. To assess the quality of the hints, and help select the most appropriate hint generation policy, we conducted a survey with experienced Alloy instructors.