Description

Advancements in neuroscience build on complex and heterogeneous data ranging from high-temporal-resolution electroencephalography (EEG) and functional/structural magnetic resonance imaging (fMRI/sMRI) to clinical records and genomic profiles. Recent developments in machine learning, particularly deep learning, have enabled the design of large-scale, robust foundation models [22] with strong generalization across data distributions, contexts, and applications. While the biggest focus has been on large language models (LLMs), which have been employed for learning from textual clinical records, several efforts have been made to develop foundation models for other neuroimaging and biosignal modalities. Emerging EEG foundation models [1, 2, 3, 4, 5] have shown promise in modelling and analysing complex EEG signals, which typically exhibit low signal-to-noise ratio and high intra- and inter-subject variability. Only very recently have attempts been made to develop such foundation models for fMRI data, leveraging large-scale cohorts to model brain activity [6, 7, 8, 19].

Most of these efforts develop models for single modalities in isolation, ignoring the complementary information that different neuroimaging and clinical data provide. As such, incorporating and analyzing multimodal data can offer a better understanding of the underlying neurological mechanisms that drive behavior and disease. Moreover, current attempts, be it uni- or multimodal, focus on learning representations from data, largely overlooking prior domain knowledge encoded in structured biomedical and/or disease-specific knowledge graphs. Structured biomedical resources, be it disease-specific knowledge graphs for Alzheimer's and epilepsy, EEG-specific ontologies, could ground model outputs in clinically explainable findings.

This PhD project will focus on the intersection of multimodal foundation models and computational neuroscience, leveraging structured, explicit domain knowledge. The aim is to develop a novel framework for constructing knowledge-grounded, interpretable, and applicable multimodal foundation models for neuroscience.

Skills required

- Master's degree (or equivalent) in computer science (machine learning, artificial intelligence), neuroscience, or related fields - Strong background in computer science, applied mathematics, and statistics, with an emphasis on machine learning (esp. deep learning) - Proficient programming skills, preferably in Python - Practical experience with machine learning/deep learning frameworks (e.g., PyTorch) - Familiarity with working/analysing EEG data, and other multimodal brain data (imaging, clinical assessments, genomic data, etc.) - Advanced proficiency in English: The candidate should be fluent in spoken and written English

Bibliography

[1] Jiang, W. et al. Large Brain Model for learning generic representations with tremendous EEG data in BCI. ICLR , 2024.

[2] Nguyen VC. et al. Balanced Latent Semantics and Signal Fidelity for EEG representation learning. ICLR TSALM 2026.

[3] Wang, Y. et al. EEGPT: Pretrained transformer for universal and reliable representation of EEG signals. In NeurIPS, 2024.

[5] Wang, Y. et al. CBraMod: A criss-cross brain foundation model for EEG decoding. 2024.

[6] Ortega Caro, J. et al. BrainLM: A foundation model for brain activity recordings. ICLR, 2024.

[7] Wang, C. et al. NeuroSTORM: Towards a general-purpose foundation model for fMRI analysis. arXiv:2506.11167, 2025.

[8] Wei, Y. et al. fMRI-LM: Towards a universal foundation model for language-aligned fMRI understanding. arXiv:2511.21760, 2025.

[9] Romano, J. D. et al. The Alzheimer's Knowledge Base: A knowledge graph for Alzheimer disease research. J Med Internet Res, 26:e46777, 2024.

[10] Pu, Y. et al. Alzheimer's disease knowledge graph enhances knowledge discovery and disease prediction. Computers in Biology and Medicine, 2025.

[11] Sahoo, S. S. et al. Epilepsy and seizure ontology: Towards an epilepsy informatics infrastructure for clinical research and patient care. J Am Med Inform Assoc, 21(1):82–89, 2014.

[12] Yang, S. et al. Knowledge graph representation of the mappings between seizure semiology and epileptogenic zones. Scientific Reports, 2026.

[13] Frishkoff, G. et al. Development of neural electromagnetic ontologies (NEMO): Ontology-based tools for representation and integration of event-related brain potentials. Nature 2009.

[14] Poldrack, R. A. et al. The Cognitive Atlas: Toward a knowledge foundation for cognitive neuroscience. Frontiers in Neuroinformatics, 5:17, 2011.

[15] Jiang, W. et al. NeuroLM: A universal multi-task foundation model for bridging the gap between language and EEG signals. In ICLR, 2025.

[16] Goswami, M. et al. MOMENT: A family of open time-series foundation models. In ICML, 2024.

[17] Feofanov, V. et al. Mantis: A generalist time series classification foundation model. 2025.

[18] Gnassounou, T. et al. Leveraging generic time series foundation models for EEG classification. arXiv:2510.27522; 2025.

[19] Tak, D. et al. BrainIAC: A generalizable foundation model for analysis of human brain MRI. Nature Neuroscience, 2026.

[20] Wang, Z. et al. Knowledge graph and its application in the study of neurological and mental disorders. Frontiers in Psychiatry, 16:1452557, 2025.

[21] Soman, K. et al. Early detection of Parkinson's disease through enriching the electronic health record using a biomedical knowledge graph. Frontiers in Medicine, 10:1081087, 2023.

[22] Zhou, X. et al. Brain foundation models: A survey on advancements in neural signal processing and brain discovery. arXiv:2503.00580, 2025.

Keywords

Time-series Foundation models , Computational Neuroscience, Multimodal machine learning, Knowledge Graphs