Probing EEG Foundation Model Attention Maps for Lightweight Interpretable Emotion Recognition
- 1Neuroengineering and Brain-Computer Interfaces Research Infrastructure, University of Patras, 26504 Patras, Greece
Abstract
Foundation Models (FMs) trained on EEG signals promise subject- and dataset-independent pipelines for brain-computer interface (BCI) applications, but have yet to consistently outperform task-specific architectures, and their large representations are costly to compute and store. As a means of advancing the understanding of the Foundation Models' inner mechanisms, we propose a lightweight alternative that exploits the self-attention mechanism of REVE, a pretrained EEG FM, to derive features similar to weighted Functional Connectivity (FC) without any fine-tuning. Inspired by kernel-based FC matrices, we aggregate the raw first-layer attention weights of a single head over the full duration of dataset trials into weighted channel-to-channel adjacency matrices. These matrices replace standard encoder representation vectors as classification features, substantially reducing computational cost. We further introduce a label-free, symmetry-based criterion, motivated by the emergence of near-symmetric attention patterns in trained encoder-decoder transformers, ranking attention heads by how much discriminative information they carry. Classical machine learning classifiers are then trained on the single most informative head's attention-derived FC matrix and achieve balanced accuracy and F1 scores on the SEED-IV emotion-recognition dataset that are competitive with several fully fine-tuned state-of-the-art EEG Foundation Models, requiring at least two orders of magnitude fewer calculations than the standard use of the encoder. Subsequent cortical region analysis revealed a significant over-representation of frontal region connections by ranking the most discriminative FC matrix connectivity edges, validating our methods as consistent with prior affective-computing literature. These results suggest that interpretable, computationally inexpensive linear probing of attention maps is a viable route to deploying EEG Foundation Models in resource-constrained settings.
Related
- LeafXNet: A Lightweight Explainable Model With Integrated Attention Mechanism for Fruit Leaf Disease Detection
- Minkowski Tensors as a Lightweight and Interpretable Representation for Three-Dimensional Morphology
- Lightweight Semantic EEG Foundation Model for Frozen Cross-Disorder Transfer
- FoME: A foundation model for EEG using adaptive temporal-lateral attention scaling
- MambaSeed: A lightweight Mamba-based hybrid vision network for tomato leaf disease recognition
BibTeX
@misc{arvanitopoulos2026probing,
title = {{Probing EEG Foundation Model Attention Maps for Lightweight Interpretable Emotion Recognition}},
author = {L. N. Arvanitopoulos and A. Dragomir and C. A. Fidas and A. Bezerianos},
howpublished = {bioRxiv},
year = {2026},
month = nov,
doi = {10.64898/2026.11.05.764629},
url = {https://www.biorxiv.org/content/10.64898/2026.11.05.764629v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Probing EEG Foundation Model Attention Maps for Lightweight Interpretable Emotion Recognition
AU - L. N. Arvanitopoulos
AU - A. Dragomir
AU - C. A. Fidas
AU - A. Bezerianos
PY - 2026
DA - 2026/11/05
PB - bioRxiv
DO - 10.64898/2026.11.05.764629
UR - https://www.biorxiv.org/content/10.64898/2026.11.05.764629v1
AB - Foundation Models (FMs) trained on EEG signals promise subject- and dataset-independent pipelines for brain-computer interface (BCI) applications, but have yet to consistently outperform task-specific architectures, and their large representations are costly to compute and store. As a means of advancing the understanding of the Foundation Models' inner mechanisms, we propose a lightweight alternative that exploits the self-attention mechanism of REVE, a pretrained EEG FM, to derive features similar to weighted Functional Connectivity (FC) without any fine-tuning. Inspired by kernel-based FC matrices, we aggregate the raw first-layer attention weights of a single head over the full duration of dataset trials into weighted channel-to-channel adjacency matrices. These matrices replace standard encoder representation vectors as classification features, substantially reducing computational cost. We further introduce a label-free, symmetry-based criterion, motivated by the emergence of near-symmetric attention patterns in trained encoder-decoder transformers, ranking attention heads by how much discriminative information they carry. Classical machine learning classifiers are then trained on the single most informative head's attention-derived FC matrix and achieve balanced accuracy and F1 scores on the SEED-IV emotion-recognition dataset that are competitive with several fully fine-tuned state-of-the-art EEG Foundation Models, requiring at least two orders of magnitude fewer calculations than the standard use of the encoder. Subsequent cortical region analysis revealed a significant over-representation of frontal region connections by ranking the most discriminative FC matrix connectivity edges, validating our methods as consistent with prior affective-computing literature. These results suggest that interpretable, computationally inexpensive linear probing of attention maps is a viable route to deploying EEG Foundation Models in resource-constrained settings.
ER -