Understanding and Correcting Low-Frequency Bias in EEG Foundation Model

2026-08-03

AI Summary

This study addresses the limited downstream performance gains of existing EEG foundation models when scaled up, which stems from a persistent low-frequency bias in their representations—arising from the coupling of the intrinsic 1/f^α spectral structure of EEG signals, neural networks’ inherent preference for low frequencies, and the use of ℓ2 reconstruction loss. To mitigate this, the authors propose FAME, a novel band-balanced masked autoencoding framework that partitions EEG signals into predefined frequency bands and performs time-frequency reconstruction independently for each band. By standardizing reconstruction targets per band and assigning equal loss weights across all bands, FAME enforces uniform supervision across the entire spectrum. Evaluated on 41 downstream tasks in the OmniEEG-Bench benchmark, FAME achieves state-of-the-art performance on 24 tasks and significantly enhances both spectral balance and transferability of learned representations.

Abstract

Increasing EEG pretraining data scale or model capacity does not consistently improve downstream performance. We identify a persistent low-frequency bias in representations learned by diverse EEG foundation models, which remains across dataset scales, model capacities, and pretraining objectives. Our analysis links this bias to the interaction between EEG's $1/f^α$-like spectral structure and neural networks' tendency to preferentially learn low-frequency components. In masked autoencoders, the $\ell_2$ reconstruction objective further amplifies this imbalance: under comparable relative reconstruction errors, high-power low-frequency components contribute disproportionately to the loss. To address this issue, we introduce FAME, a frequency-balanced masked autoencoding framework that reconstructs time--frequency activity in predefined EEG bands from masked EEG inputs. FAME independently standardizes the reconstruction targets within each band and assigns equal weight to all band-specific losses, thereby balancing supervision across the EEG spectrum. Evaluated on 41 downstream tasks in OmniEEG-Bench, FAME learns more spectrally balanced representations and achieves state-of-the-art performance on 24 of them. These results underscore the importance of balanced spectral supervision for learning transferable EEG representations.

Problem

• low-frequency bias
• EEG foundation model
• spectral imbalance
• masked autoencoding
• transferable representations

Innovation

• low-frequency bias
• EEG foundation model
• frequency-balanced reconstruction
• masked autoencoding
• spectral supervision