From Theory to Application: Fine-Tuning Large EEG Model with Real-World Stress Data

2025-05-29

AI Summary

Traditional EEG models are predominantly validated in controlled laboratory settings, limiting their applicability to real-world educational environments. Method: This study pioneers the transfer of the large-scale EEG foundation model LaBraM to authentic classroom scenarios, evaluating its capability for binary stress-state classification (resting vs. stressed) under uncontrolled conditions. Using natural resting-state EEG data from 18 graduate students, we employ supervised fine-tuning with 5-second sliding-window feature extraction and conduct systematic ablation studies—including channel reduction and random channel reordering—to assess robustness. Contributions/Results: (1) First end-to-end fine-tuning of a large EEG foundation model in an ecologically valid educational setting; (2) A paradigm shift from “model-centric” to “data-centric” BCI development; (3) Empirical validation of strong robustness against data corruption and missing channels. The model achieves a balanced accuracy of 90.47%, substantially outperforming conventional approaches while maintaining high inference efficiency.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Humans and AI: Brain-Sensing and Analysis

Application Category

• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling
• Security and Privacy: Large-scale security measurements
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Recent advancements in Large Language Models have inspired the development of foundation models across various domains. In this study, we evaluate the efficacy of Large EEG Models (LEMs) by fine-tuning LaBraM, a state-of-the-art foundation EEG model, on a real-world stress classification dataset collected in a graduate classroom. Unlike previous studies that primarily evaluate LEMs using data from controlled clinical settings, our work assesses their applicability to real-world environments. We train a binary classifier that distinguishes between normal and elevated stress states using resting-state EEG data recorded from 18 graduate students during a class session. The best-performing fine-tuned model achieves a balanced accuracy of 90.47% with a 5-second window, significantly outperforming traditional stress classifiers in both accuracy and inference efficiency. We further evaluate the robustness of the fine-tuned LEM under random data shuffling and reduced channel counts. These results demonstrate the capability of LEMs to effectively process real-world EEG data and highlight their potential to revolutionize brain-computer interface applications by shifting the focus from model-centric to data-centric design.

Problem

• Evaluating Large EEG Models for real-world stress classification
• Assessing model robustness with shuffled data and fewer channels
• Transitioning EEG analysis from clinical to real-world settings

Innovation

• Fine-tuned LaBraM model with real-world EEG data
• Achieved 90.47% accuracy in stress classification
• Evaluated robustness under data shuffling and reduced channels