Reducing Interdataset Covariate Shift in Sleep EEG of Traumatic Brain Injury Using Transfer Euclidean Alignment

This project introduces Transfer Euclidean Alignment (TEA), a transfer learning technique that reduces variability across sleep EEG datasets to improve the generalizability of machine learning models. Traditional models often fail when applied to new datasets due to covariate shifts, especially in biomedical applications where data are scarce. TEA aligns EEG data distributions across both humans and mice, enabling cross-dataset and cross-species learning for traumatic brain injury (TBI) detection. Evaluations with EEGNet and other models show notable accuracy gains — averaging +14.4% in human datasets and +5.5% in mouse-to-human transfer — highlighting TEA’s potential to boost robustness and advance clinical applications of sleep EEG.

Publications:


Full paper

Project information

  • Category: Future Healthcare, Health State Estimation, and Personal Health Models
  • Contact Person: Hung Cao
  • Reducing Interdataset Covariate Shift in Sleep EEG of Traumatic Brain Injury Using Transfer Euclidean Alignment

info@futurehealth.uci.edu

© Copyright 2021 UCI Institute for Future Health - All Rights Reserved