EEG-JEPA: Structured Latent Prediction for EEG Foundation Models

2026-07-31

AI Summary

This work addresses the limitations of traditional EEG foundation models, which rely on masked waveform reconstruction and are thus highly susceptible to noise and artifacts, hindering the learning of transferable neural representations. To overcome this, the authors propose EEG-JEPA, a novel framework that eschews raw signal reconstruction in favor of modeling neural activity through structured latent variable prediction. The method innovatively constructs prediction targets along three dimensions: content semantics, neurotopology-informed multi-scale electrode-time masking (N-MET), and target depth. Coupled with an exponential moving average target encoder and hierarchical supervision, EEG-JEPA enables end-to-end inference of latent neural states. Evaluated on EEG-FM-Bench, the model achieves a frozen 14-task macro-average accuracy of 52.94% and a fully fine-tuned 9-task average accuracy of 70.65%, significantly outperforming existing approaches.

Abstract

Electroencephalography (EEG) foundation models aim to learn reusable representations from large-scale unlabeled recordings. A common pretraining strategy is masked waveform reconstruction, but applying supervision directly to noisy EEG may encourage models to recover predictable background activity, acquisition effects, and artifacts rather than neural structure that transfers across tasks. This raises a central question: what should an EEG foundation model predict to learn transferable representations? We introduce EEG-JEPA a structured latent-prediction framework for EEG foundation modeling. Rather than reconstructing masked voltage samples, a masked context encoder and predictor infer contextual latent states produced by an exponential-moving-average target encoder that observes the complete input. EEG-JEPA organizes target design along three complementary dimensions: target content specifies what representation is predicted, target support specifies where prediction occurs over structured electrode--time regions through Neurotopology-Aware Multi-scale Electrode-Temporal Masking (N-MET), and target depth specifies at which encoder layers supervision is applied. Together, these designs shift EEG pretraining from recovering missing measurements to inferring latent states from structured electrode--time context. We evaluate EEG-JEPA through controlled objective comparisons, frozen multitask transfer, and full fine-tuning. Under the same backbone, pretraining corpus, and training duration, EEG-JEPA improves the 14-task frozen macro balanced accuracy from 40.49% to 50.42% over CBraMod-style masked waveform reconstruction. Multi-source continuation further raises this result to 52.94%, the highest average among the EEG foundation models evaluated on EEG-FM-Bench. Under protocol-matched full fine-tuning, EEG-JEPA also improves the nine-task average balanced accuracy from 68.98% to 70.65%.

Problem

• EEG foundation models
• masked waveform reconstruction
• transferable representations
• latent prediction
• pretraining objective

Innovation

• structured latent prediction
• EEG foundation model
• Neurotopology-Aware Multi-scale Electrode-Temporal Masking
• target encoder
• masked context prediction