# BandVQ: Band-Wise Vector-Quantized EEG Foundation Model

arXiv: https://arxiv.org/abs/2605.24921 · HTML: https://mlsanity.com/paper/2605.24921/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 5 of its 16 references that are in the map.
- First posted on arXiv 24 May 2026.

## Builds on (most cited first, 5 of 5)

- [Neural Discrete Representation Learning](https://mlsanity.com/paper/1711.00937/) (2017)
- [Large Brain Model for Learning Generic Representations with Tremendous EEG Data in BCI](https://mlsanity.com/paper/2405.18765/) (2024)
- [BENDR: using transformers and a contrastive self-supervised learning task to learn from massive amounts of EEG data](https://mlsanity.com/paper/2101.12037/) (2021)
- [CBraMod: A Criss-Cross Brain Foundation Model for EEG Decoding](https://mlsanity.com/paper/2412.07236/) (2024)
- [SingLEM: Single-Channel Large EEG Model](https://mlsanity.com/paper/2509.17920/) (2025)

## Built on by (most cited first, 1 of 1)

- [SpecBraM: What Should an EEG Foundation Model Predict? Masked Band-Power Prediction versus Waveform Reconstruction](https://mlsanity.com/paper/2610.07484/) (2026)

## Abstract (arXiv)

A central challenge in electroencephalography (EEG) foundation modeling is learning transferable representations across recordings with diverse tasks, montages, references, and spectral characteristics. Existing masked modeling approaches often rely on broadband continuous patches or a single discrete representation, which may underrepresent frequency-specific activity. This paper proposes BandVQ, a band-wise vector-quantized EEG foundation model that decomposes EEG into delta, theta, alpha, beta, and gamma bands, trains an independent VQ-VAE tokenizer for each band, and pretrains a shared Transformer encoder on the resulting discrete VQ code indices. The encoder uses masked code tokens, quantized absolute log-power tokens, channel and temporal embeddings, and metadata prefix tokens representing reference, band, task family, and phase. Region-based masking is also introduced to reduce the trivial reconstruction of spatially adjacent electrodes. The model is pretrained on 71 public EEG corpora comprising over 9,200 subjects and 357,000 single-channel hours and evaluated on six subject-independent classification datasets. Under the current evaluation setting, the proposed model achieves strong transfer performance, with the highest reported results on three cognitive tasks and competitive performance on three motor imagery tasks.