DISA

When? Friday March 1st 14-16
Where? Onsite: B1009 at Linnaeus University in Växjö and online
Registration: Please sign up for the PhD-seminar via this link https://forms.gle/1KCLvBZQph4Vs1cR7 by February 28th (especially important if you plan on attending onsite so we have fika for everyone)

14.00-14.10 Welcome and practical information from Welf Löwe
14.10-14.55 Presentation and discussion: Leveraging machine learning and large language models to map source code as natural language to architectural modules. – Nils Johansson, Industry PhD-student at Volvo CE
14.55 – 15.05 Coffee break
15.05 – 15.50 Presentation and discussion – Towards a Probabilistic Interactive Video Training Tool for Emergency Medical Services – Romain Herault, PhD-student at LNU
15.50 -16.00 Sum up and plan for our next seminar on April 5th

Abstracts

Leveraging machine learning and large language models to map source code as natural language to architectural modules – Nils Johansson, Industry PhD-student at Volvo CE

It is not uncommon for software developing companies to be in a situation where an old software system is maintained for years on end. During this time or when at first development, it may suffer architectural deterioration or drift. However, refactoring a software system to a new architecture is a costly process that requires great effort and commitment. When refactoring using a big-design-up-front approach, a new design of the software is initially created. Alternatively, a reference architecture is used. When a new architecture is obtained, begins a process of assigning code or functionality to each module of the new architecture. Instead of painstakingly trying to analyze the old code to decide which functionality should be assigned to certain modules, it would be beneficial to perform this mapping automatically. In this study, this activity is attempted to be solved by formalizing it as a natural language problem using machine learning and specifically large-language models. Descriptions of modules that include the “roles and responsibilities” are used to constitute classes. Source code snippets from the old software are then classified to suitable classes(modules) from the new architecture, it is thereby a text classification problem. An important aspect to explore is how detailed descriptions that are needed. Likely, with extensive descriptions of roles and responsibilities the classification will be more accurate, but the situation is not as applicable for the real use case.

Towards a Probabilistic Interactive Video Training Tool for Emergency Medical Services – Romain Herault, PhD-student at LNU

Emergency Medical Services (EMS) professionals undergo continuous training, crucial for handling high-pressure situations. Innovative approaches are necessary to enhance the effectiveness of EMS training, considering the time constraint these professionals are under. Probabilistic interactive video training is a promising avenue. It employs interactive web-based platforms to create immersive learning experiences via a standard web browser. Personalized training using probabilistic models tailors the training to individual trainees’ needs and enhances engagement. The created probabilistic models simulate realistic emergency scenarios that foster the development of robust decision-making skills under uncertain and time-critical conditions. The research aims to analyze the effectiveness of interactive video-based (both regular and 360-degree) training and explore its potential as an innovative approach to enhance EMS training using the Technology Acceptance Model (TAM) and a mix of interviews and focus groups with police and ambulance students.