Cross-Cohort Spectral-Temporal Dissociation in Frozen EEG Foundation-Model Representations
Researchers explored whether frozen representations from five EEG foundation models can decode long-range temporal correlations in brain activity. They tested these models on two datasets, CAUEEG and BrainLat, using various methods to analyze the data. The results showed that some models, such as CBraMod and BIOT, were able to decode aperiodic patterns in both cohorts, but decoding of alpha-band amplitude envelope was cohort-dependent. This study provides new insights into th
Researchers explored whether frozen representations from five EEG foundation models can decode long-range temporal correlations in brain activity. They tested these models on two datasets, CAUEEG and BrainLat, using various methods to analyze the data. The results showed that some models, such as CBraMod and BIOT, were able to decode aperiodic patterns in both cohorts, but decoding of alpha-band amplitude envelope was cohort-dependent. This study provides new insights into the representational capabilities of EEG foundation models.
---
Why it matters: This research matters because it helps engineers understand the strengths and limitations of different EEG foundation models, which can inform the development of more accurate brain-computer interfaces and other applications.
Source: https://arxiv.org/abs/2607.24834
This article was originally published at: https://arxiv.org/abs/2607.24834