Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

Siwen Wang, Shitou Zhang, Wan-Lin Chen, Dung Truong, Tzyy-Ping Jung

TL;DR

This study demonstrates that a Large EEG Model (LaBraM) can be effectively fine-tuned on a real-world classroom stress EEG dataset, achieving a single-window 5-second binary stress classifier with high accuracy. The authors systematically preprocess, augment, and segment resting-state EEG from 18 students, then fine-tune a 5.8M-parameter LEM, revealing strong robustness to data splits, significant gains from pretraining, and sensitivity to channel count. The findings support a data-centric BCI paradigm and highlight practical considerations for real-world deployment, including inference efficiency and the need for interpretability. Overall, the work extends LEM applicability beyond controlled clinical settings and provides a benchmark for real-world stress detection with EEG, with implications for near-real-time BCI systems.

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.

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

TL;DR

This study demonstrates that a Large EEG Model (LaBraM) can be effectively fine-tuned on a real-world classroom stress EEG dataset, achieving a single-window 5-second binary stress classifier with high accuracy. The authors systematically preprocess, augment, and segment resting-state EEG from 18 students, then fine-tune a 5.8M-parameter LEM, revealing strong robustness to data splits, significant gains from pretraining, and sensitivity to channel count. The findings support a data-centric BCI paradigm and highlight practical considerations for real-world deployment, including inference efficiency and the need for interpretability. Overall, the work extends LEM applicability beyond controlled clinical settings and provides a benchmark for real-world stress detection with EEG, with implications for near-real-time BCI systems.

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.

Paper Structure

This paper contains 11 sections, 4 figures, 2 tables.

Table of Contents

  1. INTRODUCTION
  2. Method
  3. Large EEG Model
  4. Downstream Fine-tuning Dataset
  5. Participants
  6. Experimental Design
  7. Pre-processing
  8. Further Data Processing Before Fine-tuning
  9. LEM Fine-tuning
  10. Experiments and Results
  11. Discussion

Figures (4)

  • Figure 1: Further Data Processing Pipeline
  • Figure 2: Finetuning results using 4 different data splitting seeds.
  • Figure 3: Fine-tuning results with pre-train v.s. without pre-train. Average results across different seeds are highlighted in darker colors
  • Figure 4: Fine-tuning results with different channel counts. Average results across different seeds are highlighted in darker colors