Kategorie: ‘Theses’
Frequent Learner Errors: NLP Insights into Automated Grading
This thesis aims to explore the realm of Natural Language Processing (NLP) in the context of automated grading systems, focusing specifically on identifying and analyzing frequent learner mistakes. With the increasing integration of technology in education, automated grading systems have gained prominence, but they often lack nuanced understanding and feedback provision. This research endeavors to enhance the efficacy of such a system in use at the institute by employing NLP techniques to dissect learner errors and provide more tailored feedback.
Automated Data Processing and Knowledge Discovery for Time Series Data utilizing Large Language Models
Large language models (LLMs) have proven the ability to assist diverse users in conducting a variety of individual tasks via intuitive and natural conversations. This thesis discusses a utilization of LLMs to perform (semi-)automated data processing and analyses on time series data. One major goal is to reduce expertise-related dependencies, allowing more people to manipulate data and gain beneficial insights.
Explainable Data – Trust, Transparency and Bias Mitigation in ML
This bachelor thesis aims to delve into the critical intersection of trust, transparency, and bias mitigation in machine learning (ML) systems through the lens of explainable data. The proliferation of ML algorithms across various domains has underscored the importance of understanding how these systems make decisions, especially when they impact individuals or societal outcomes.
Adaptive Semantics for Generic Latex Expressions in the Context of Educational Resources
This thesis aims at developing a pipeline for automatic translation of generic latex expression into domain specific, processable, languages.
Need for Speed: Evaluating Feedback Latency in Psychomotor Learning
The bachelor thesis aims to contribute to the ongoing research project MILKI-PSY, which is centered around advancing Self-Regulated Learning (SRL) in psychomotor training through multimodal immersive mentoring. In the context of the project, we explore how various calculation mechanisms and hardware decision impact the latency of feedback in psychomotor learning.
An Empirical Study of Open Source Large Language Models (OSLLMs)
Open Source Software (OSS) revolutionised the computing world about three decades ago. One of the principles of OSS guarantees software developers, companies, researchers, and students the freedom to change and improve the software. Characterised by active community involvement (bazaar-style software development), OSS development has produced category-killer Operating Systems (e.g., Debian, Ubuntu) and applications (e.g., the Apache HTTP Server, Firefox).
The computer science community is now riding another revolution called the Large Language Models (LLMs) revolution. Various variants of LLMs (Commercial and Open Source) come with billions of parameters that developers can fine-tune to control how the system generates text (tokens). Commercial LLMs (e.g., ChatGPT) come with a copyright and are expensive to deploy and use. They have also been criticised for their hallucination, lack of transparency, and the potential for monopolisation by big corporations.
Building a Social Bot for Self-Regulated Learning: Leveraging xAPI Statements, Visualizing Progress, and Recommending Learning Materials
This thesis aims to develop a social bot designed to enhance self-regulated learning experiences by leveraging xAPI statements to track learning progress. The system will employ the “GRETA Kompetenzbilanz” model to visualize the learner’s progress comprehensively, recommend suitable learning materials, and facilitate reflective practices. The project also explores the potential integration of FAQ handling and considers the implementation of MLOps for continuous improvement through retraining of recommendation models.
Algorithmic Approaches to Overlapping Community Detection – Multiplex Network Compatibility and a Chatbot Environment
Psychometrics of AI Dictators: Creating and Analyzing LLM Criminal Personas
Psychometric inventories provide a means to analyze the expression of a variety of human traits. When applied on specific sets of people, they may show shared characteristics that could contribute to their behavior. Common traits can be significant for criminal individuals, as they could be used to discern early warning signs or identifiers. For example, certain profiles of the Dark Triad psychometric test which is composed of psychopathy, machiavellianism, and narcissism, and self-control, have been related to antisocial and criminal externalizing outcomes. In this thesis, we explore the idea of applying psychometrics to generative LLMs that impersonate well-known criminals and dictators to circumvent problems of limited/restricted access to actual, and often historic, people.