Problem Description

Biosignals, ranging from brainwaves and eye tracking to body movements, provide a comprehensive real-time insight into human physiology. Traditionally, accurately and efficiently interpreting these signals has demanded specialised expertise and has often been hindered by processing bottlenecks in terms of volume and speed. Meanwhile, large language models (LLMs) have transformed the way we process complex, sequential data, showcasing an impressive capacity to recognise intricate patterns and relationships in natural language. This raises a compelling question: Can LLMs extend their powerful analytical capabilities beyond text to effectively interpret non-linguistic time series biosignals? There is a significant knowledge gap here. While other AI models have addressed biosignals, the unique strengths of transformer-based LLMs — their contextual understanding and capacity to handle long dependencies — have not been extensively explored in relation to this data type. Consider, for example, the potential application of biosignals (such as eye tracking or neural activity) in economic decision-making to understand real-time user engagement or stress during financial transactions. Without clarity on LLMs' suitability for this task, vital opportunities for innovation in areas requiring rapid, nuanced biosignal interpretation will be missed.

Goal of the Thesis

The aim of this thesis is to systematically explore the potential of Large Language Models (LLMs) for general biosignal processing and interpretation. The key goals are:

• Systematic literature analysis & conceptualisation: Reviewing existing research on the application of LLMs, or similar models, to biosignal analysis. This will involve identifying current methods, successes and limitations, as well as fundamental questions regarding the ability of LLMs to process and extract meaningful insights from diverse biosignal data.
• Preliminary experimental investigation (pre-study): Conduct initial small-scale tests using selected LLMs and representative biosignals. This pre-study will provide practical insights into the feasibility and challenges of applying LLMs in this field and inform the design of the main experiment.
• Comprehensive experimental design: Developing a detailed and robust plan for a thorough study to evaluate LLMs' capabilities in biosignal processing, including methodologies, datasets and performance metrics.

This thesis ultimately aims to provide a foundational understanding of, and clear direction for, future research at the intersection of large language models and general biosignal analysis. It will highlight the potential for new applications in diverse fields, particularly those involving human behaviour and decision-making.

Requirements

• You should have a basic knowledge of scientific work, including systematic literature analysis and experimental design, or be willing to undertake self-study.
• Having a basic understanding of LLMs, prompting and fine-tuning methods, and being familiar with data preparation mechanisms, is advantageous.

Starting Literature

• Ariely, D., & Berns, G. S. (2010). Neuromarketing: The hope and hype of neuroimaging in business. Nature Reviews Neuroscience, (4), 284–292. 
• Iranmanesh, M., Gunaratnege, S. M., Ghobakhloo, M., Foroughi, B., Yadegaridehkordi, E., & Annamalai, N. (2024). Determinants of intention to use ChatGPT for obtaining shoppinginformation. Journal of Marketing Theory and Practice
• Pfeiffer, J., Pfeiffer, T., Meißner, M., & Weiß, E. (2020). Eye-tracking-based classification of information search behavior using machine learning: Evidence from experiments in physicalshops and virtual reality shopping environments. Information Systems Research ,(3), 675–691.
• Schultz, T., & Maedche, A. (2023). Biosignals meet adaptive systems. SN Applied Sciences, 5(9), 234.
• Weiß, T., & Pfeiffer, J. (2024). Consumer decisions in virtual commerce: Predict good help-timing based on cognitive load. Journal of Neuroscience, Psychology, and Economics, 17(2), 119.