Description

Large language models (LLMs) represent meaning using statistical patterns learned from large-scale text data. However, human semantic representations are not necessarily identical to these average model-derived structures: different individuals may organize concepts in subtly different ways. Understanding whether brain activity can reveal such individual semantic organization is an important question for cognitive neuroscience, neurotechnology, and future human–AI interaction.

This project explores a new direction for brain–computer interfaces: using neural responses not only to detect intentions, but also to estimate how object meaning is represented in the brain and to enrich human–AI interaction with this information. The student will work with open EEG, iEEG and other neuroimaging datasets in which participants viewed images of objects or words. Representational similarity analysis and other multivariate approaches, combined with behavioral data analysis, will be used to construct brain-derived semantic maps, and then translate them into a format usable by an AI agent.

The project will be primarily computational and will involve analysis of existing open datasets. Depending on the student’s interests and skills, the thesis may focus more on neuroscience analysis, LLM interaction, or the bridge between both.

We anticipate approximately 2–3 months for literature review and dataset preparation, 3–4 months for data analysis and the pipeline implementation, and 2 month for validation, and thesis writing. The starting date can be discussed.

The student will have the opportunity to participate in the following:

• Conduct a literature review on semantic representations, RSA, and brain–AI interfaces.
• Work with open EEG/fMRI/iEEG datasets.
• Preprocess and analyze neuroimaging data using Python.
• Develop LLM-readable semantic context files or retrieval structures.
• Potentially contribute to a publication based on the project.

Desired background:

• Interest in cognitive neuroscience and brain–computer interfaces.
• Solid Python programming skills.
• Interest in AI, large language models, and LoRA-based model adaptation.
• Experience with EEG, neuroimaging, or statistics is helpful but not required.

To apply, please send a CV, a short statement of motivation, and your availability, including preferred starting date, expected duration, and hours per week available for the project.

Contact: Nikolai Syrov

Email: Nikolai.Syrov@ruhr-uni-bochum.de