B[FM]$^2$: Brain Foundation Model via Flow Matching with SplitUNet

2026-06-18

AI Summary

This work addresses a critical limitation in existing EEG foundation models, which rely on discretization procedures that disrupt continuous neural rhythms and discard fine-grained temporal dynamics. To overcome this, the authors propose a continuous-time flow-matching pretraining paradigm that operates directly on raw EEG signals—eliminating the need for chunking, tokenization, or masking. They further introduce a SplitUNet architecture that decouples the modeling of 1D temporal dynamics from electrode topological structure, effectively handling the inherent asymmetry between EEG’s spatial and temporal dimensions. Remarkably, using only approximately 307 hours of data—roughly 1/30th of what current methods require—the model achieves new state-of-the-art performance on seven out of nine standard downstream classification tasks. Moreover, synthetic EEG generated by the model was deemed indistinguishable from real recordings by neurologists (Cohen’s κ = −0.096).

Abstract

EEG foundation models can learn generalizable representations from large-scale EEG corpora to enable single-backbone transfer across diverse clinical and brain-computer interface tasks. Existing models typically discretize the continuous multi-channel EEG waveform into patches or codebook tokens and train a transformer with masked self-supervision. Recognizing that this discretization fragments continuous brain rhythms and obscures fine-grained temporal dynamics, we present B[FM]$^2$(Brain Foundation Model via Flow Matching), whose inductive bias aligns with the data by pretraining directly on the raw signal using continuous-time flow matching without patches, tokenization, or masking. However, multi-channel EEG signals pose an architectural challenge for flow matching: time is densely sampled and highly autocorrelated (thousands of timepoints), while the electrode axis is short (tens of channels) at distinct scalp positions. To address this time-electrode asymmetry, we introduce SplitUNet, a velocity network that factorizes each block into separate 1D temporal and 1D electrode convolutions and downsamples only along time, preserving electrode topology throughout the hierarchy. B[FM]$^2$ sets a new state of the art on 7 of 9 standard downstream EEG classification tasks, using a pretraining budget of only 36,895 segments ($\approx$ 307h), 1-2 orders of magnitude ($\approx$ 30x) less than required by existing EEG foundation models. Further, it generates synthetic EEGs that two board-certified neurologists cannot distinguish from brain data (Cohen's $κ=$ -0.096). https://jd730.github.io/projects/BFM2

Problem

• EEG foundation model
• continuous-time modeling
• temporal dynamics
• signal discretization
• time-electrode asymmetry

Innovation

• Flow Matching
• SplitUNet
• EEG Foundation Model
• Continuous-time Modeling
• Electrode Topology Preservation