THD-BAR: Topology Hierarchical Derived Brain Autoregressive Modeling for EEG Generic Representations

TL;DR

THD-BAR tackles the challenge of learning universal EEG representations by incorporating brain topology into autoregressive modeling. It introduces the Brain Topology Hierarchy (BTH), a topology-aware multi-scale spatial ordering, and the THVQ-VAE tokenizer to produce discrete tokens, combined with a next-scale-time autoregressive BAR module. Pre-trained on 17 diverse EEG datasets and validated on 10 downstream tasks, THD-BAR demonstrates superior generalization and robust spatio-temporal modeling compared to prior autoregressive methods. This framework advances EEG foundation modeling and enables more accurate cross-task EEG analysis with potential for multi-modal extension.

Abstract

Large-scale pre-trained models hold significant potential for learning universal EEG representations. However, most existing methods, particularly autoregressive (AR) frameworks, primarily rely on straightforward temporal sequencing of multi-channel EEG data, which fails to capture the rich physiological characteristics inherent to EEG signals. Moreover, their time-centered modeling approach also limits the effective representation of the dynamic spatial topology of brain activity. To address these challenges and fully exploit the potential of large-scale EEG models, we propose a novel Topology Hierarchical Derived Brain Autoregressive Modeling (THD-BAR) for EEG generic representations. The core innovation of THD-BAR lies in the introduction of the Brain Topology Hierarchy (BTH), which establishes a multi-scale spatial order for EEG channels. This hierarchical structure enables a redefinition of autoregressive learning as a "next-scale-time prediction" problem, effectively capturing both spatial and temporal dynamics. Based on BTH, we design a Topology-Hierarchical Vector Quantized-Variational Autoencoder (THVQ-VAE) for multi-scale tokenization and develop an enhanced Brain Autoregressive (BAR) module with specialized masking strategies for prediction. Through extensive large-scale pre-training on 17 datasets, followed by rigorous validation on 10 downstream datasets spanning 5 distinct tasks, THD-BAR consistently outperforms existing methods. These results highlight the superior generalization and modeling capabilities of our proposed approach.

Figures (13)

• Figure 1: The overall performance comparison on 10 EEG datasets.
• Figure 2: Conceptual comparison of autoregressive prediction strategies.
• Figure 3: The three-stage pipeline of the THD-BAR framework. (1) An THVQ-VAE tokenizes unlabeled EEG data. (2) The BAR model undergoes autoregressive pre-training using these tokens. (3) The pre-trained BAR model is fine-tuned for specific downstream tasks with labeled data.
• Figure 4: Topology-Hierarchical Vector Quantized-Variational Autoencoder (THVQ-VAE).
• Figure 5: A BAR causal transformer is trained via "next-scale-time prediction".