Adnan KhanAdnanProduct & AI
San Francisco Bay Area
Get in touch

Source note2 min read

MSBraM: A Multi-scale Self-supervised Brain Foundation Model for Hierarchical EEG Dynamics Learning

Source

arXiv

arxiv.org/abs/2607.21402v1

Read the source

Source

Plain-Language Brief

The authors propose a new model named MSBraM to analyze electroencephalogram (EEG) signals more effectively. Existing methods struggle with the multi-scale temporal structure of EEG data, which includes both local neural patterns and long-range dependencies. The proposed model learns hierarchical EEG representations in a two-stage process: first by discretizing raw EEG signals into semantic codes at different temporal resolutions, then pretraining the model to predict masked codes using a curriculum multi-scale masking strategy. The model was pretrained on over 2400 hours of EEG data and tested across ten downstream tasks on twelve public datasets. Results show that MSBraM outperforms other state-of-the-art pretrained models, indicating the importance of explicitly modeling multi-scale temporal dynamics for effective EEG analysis.

What the Document Says

  • The authors propose a model named MSBraM for EEG analysis.
  • Existing methods have difficulties capturing the multi-scale temporal structure of EEG signals.
  • MSBraM learns hierarchical EEG representations in two stages: vector-quantized reconstruction and pretraining to predict masked codes using a curriculum multi-scale masking strategy.
  • The model was pretrained on over 2400 hours of EEG data.
  • MSBraM was tested across ten downstream tasks on twelve public datasets.
  • Results show that MSBraM outperforms other state-of-the-art pretrained models.

Key Concepts Mentioned

  • EEG analysis
  • self-supervised foundation models
  • Multi-Scale self-supervised Brain foundation Model (MSBraM)
  • multi-scale temporal structure
  • vector-quantized reconstruction
  • curriculum multi-scale masking strategy

Claims Made by the Document

  • Existing approaches struggle to capture the inherently multi-scale temporal structure of EEG signals.
  • MSBraM achieves superior performance on other state-of-the-art pretrained models.

Evidence Used

  • The text states that existing methods have difficulties capturing the multi-scale temporal structure of EEG signals.
  • The text provides results showing that MSBraM outperforms other state-of-the-art pretrained models.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Self-supervised foundation models have recently shown strong potential for electroencephalogram (EEG)-based analysis. However, existing approaches struggle to capture the inherently multi-scale temporal structure of EEG signals, where local neural patterns and long-range dependencies jointly encode task-relevant information. This limitation hampers cross-scale representation learning and generalization across diverse downstream tasks. To address this challenge, we propose MSBraM, a Multi-Scale self-supervised Brain foundation Model designed to learn hierarchical EEG representations. MSBraM follows a two-stage pretraining framework. First, a multi-scale neural tokenizer discretizes raw EEG signals into semantic codes at different temporal resolutions via vector-quantized reconstruction. Second, the model is pretrained to predict masked codes using a curriculum multi-scale masking strategy, progressively integrating fine-grained local patterns with global temporal context. We pretrain MSBraM on over 2,400 hours of EEG data and evaluate it across 10 downstream tasks on 12 public datasets. Extensive experiments show that MSBraM achieves superior performance on other state-of-the-art pretrained models, demonstrating strong generalization and transferability. These results indicate that explicitly modeling multi-scale temporal dynamics is critical for effective EEG foundation models.