When two people talk, their brains synchronize in complex ways. But can we actually decode anything about conversation participants from brain activity? This project uses machine learning on simultaneous brain recordings (hyperscanning EEG) from conversing pairs to rigorously test whether neural patterns generalize across different conversational partners.
Previous pilot work (Googel Summer of Code 2025, N=8 pairs) achieved 94% accuracy in decoding both conversational roles and participant gender using CEBRA (Contrastive Embedding for Behavioral and Neural Analysis). With our expanded dataset of 40+ pairs, we are looking to build mathematical understanding of which features of neural data inform successful brain-to-brain mapping (explainable AI), and to compare/contrast validation models for topological mapping.
Recent research challenges the “deficit model” of social communication differences. Crompton et al. (2025, Nature Human Behaviour) showed that communication breakdowns depend more on neurotype mismatch than individual deficits; but these studies used only behavioral measures. Your work establishes the neural foundation for understanding how brain-to-brain coordination enables (or impairs) communication, with implications for designing assistive technologies.
Total project length: 350 hours (12 weeks, full-time).
Please use this link to access the test for this project.
Essential: Python (NumPy, SciPy), basic signal processing (filtering, FFT) Preferred: Experience with time-series data (including neural), MNE-Python
Intermediate
Please DO NOT contact mentors directly by email. Questions should instead be directed to ml4-sci@cern.ch which is forwarded to mentors. To submit your proposal, CV, and test task solutions, please use this Google form.