Thesis (Master): Beyond LLMs: Rethinking Foundation Models for Time Series Forecasting

<![[<div><h1>Aufgaben:</h1>You apply your academic knowledge to the systematic literature review of structural differences between language and time series data within Transformer-based architectures
Building on this, you critically assess which Transformer components-such as tokenization, positional encoding, and attention mechanisms-can be successfully transferred from LLMs to time series modeling and which require adaptation
Subsequently, you conduct a targeted practical investigation of a specific finding, for example by analyzing and redesigning components like embedding layers or positional encodings, or by exploring fine-tuning strategies for time series foundation models
To validate your hypotheses, you design and execute experiments that evaluate the impact of the identified architectural modifications
Where appropriate, you extend your research to the energy systems domain, using real-world datasets from ongoing projects to test the applicabil...