Researchers have developed a parameter-efficient self-supervised adaptation method for EEG foundation models (EEG-FM) to improve generalization across diverse clinical datasets. This approach, which updates only 9% of model parameters, was evaluated on two state-of-the-art models and three clinical EEG datasets for abnormality detection, event classification, and seizure detection. The findings indicate consistent performance gains over linear probing, with peak performance achievable using only 20-50% of unlabeled data under a fixed compute budget. Notably, performance was found to be invariant to patient count when the total window count remained fixed, suggesting that overall temporal window diversity is the key factor. AI
IMPACT Enables more efficient deployment of EEG foundation models in resource-constrained clinical settings, reducing computational overhead and data burden.
RANK_REASON This is a research paper detailing a new method for adapting existing models. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
