EEG models improved by separating subject traits from brain signals
DEFT: Disentanglement-Enhanced Fine-Tuning for EEG Foundation Models
Machine Learning
Summary
EEG models often mix up individual differences with brain activity related to tasks, making it harder to analyze signals across people. The authors propose DEFT, a method that splits the model’s understanding into two parts: one capturing task-specific brain signals and the other capturing individual traits. By focusing on the task-related signals for predictions, their method improves performance on multiple EEG datasets and models. This approach helps the computer better understand what brain signals mean across different people.
What this means in practice
- •For medical device developers: Improve brain signal interpretation in EEG-based diagnostic tools by reducing individual subject variability in model embeddings.$Commercial implications: Enables more accurate and generalizable EEG diagnostics devices that adapt to different patients.
- •For brain-computer interface engineers: Enhance cross-subject performance of EEG classifiers used in brain-computer interfaces by disentangling subject style from task content.
Authors
Yuqiu Deng, Muyun Jiang, Zongpeng Zhang, Mingqing Xiao, Yansen Wang, Peiliang Gong, Xinliang Zhou, Yi Ding, Ziyu Jia, Cuntai Guan, Chenyu Liu
Abstract
EEG foundation models (EFMs) have improved EEG representation learning by transferring knowledge from large-scale multi-subject pretraining to downstream tasks. However, their pretrained embeddings often retain substantial subject-related variation, placing an additional burden on the downstream classifier, especially in cross-subject settings. To address this issue, we propose Disentangled EFM Fine-Tuning (DEFT), a plug-and-play adapter for EFM fine-tuning. DEFT encourages branch disentanglement of pretrained tokens into task-oriented content and subject-consistent style, then performs downstream prediction from the content branch. It instantiates this factorize-before-classify design with recursive Gaussian refinement, temporal-variance regularization, subject-aware contrastive learning, and reconstruction. Experiments across multiple EEG datasets and four EFM backbones show that DEFT improves matched fine-tuning baselines. Further ablations and analyses support the intended content/style branch organization.