BrainPro: Towards Large-scale Brain State-aware EEG Representation Learning

Electroencephalography (EEG) is a non-invasive technique for recording brain electrical activity, widely used in brain-computer interface (BCI) and healthcare. Recent EEG foundation models trained on large-scale datasets have shown improved performance and generalizability over traditional decoding methods, yet significant challenges remain. Existing models often fail to explicitly capture channel-to-channel and region-to-region interactions, which are critical sources of information inherently encoded in EEG signals. Due to varying channel configurations across datasets, they either approximate spatial structure with self-attention or restrict training to a limited set of common channels, sacrificing flexibility and effectiveness. Moreover, although EEG datasets reflect diverse brain states such as emotion, motor, and others, current models rarely learn state-aware representations during self-supervised pre-training. To address these gaps, we propose BrainPro, a large EEG model that introduces a retrieval-based spatial learning block to flexibly capture channel-and region-level interactions across varying electrode layouts, and a brain state-decoupling block that enables state-aware representation learning through parallel encoders with decoupling and region-aware reconstruction losses. This design allows BrainPro to adapt seamlessly to diverse tasks and hardware settings. Pre-trained on an extensive EEG corpus, BrainPro achieves state-of-the-art performance and robust generalization across nine public BCI datasets. Our codes and the pre-trained weights will be released. Electroencephalography (EEG) provides a non-invasive and cost-effective window into large-scale brain activity, supporting a wide range of applications in brain-computer interfaces (BCIs), cognitive neuroscience, and clinical neurotechnology. Despite its potential, EEG data are notoriously challenging to model due to low signal-to-noise ratio, non-stationarity, and variability across subjects and recording setups (Schalk et al., 2024; Wang et al., 2024b). Traditional EEG analysis methods relied on handcrafted features (e.g., spectral power, connectivity measures) tailored to specific tasks, but these approaches are labor-intensive and lack generalizability. With the advent of deep learning, supervised convolutional and recurrent neural networks have been applied to tasks such as motor imagery classification, sleep staging, and emotion recognition, but their reliance on large amounts of labeled data limits scalability (Jiang et al., 2024; Wang et al., 2025). Inspired by the success of foundation models in language and vision (Devlin et al., 2018; He et al., 2022; Radford et al., 2021), researchers have recently proposed EEG foundation models (EFMs) trained on large-scale, unlabeled datasets using self-supervised learning (Zhou et al., 2025). These models aim to learn generalizable representations that transfer across tasks and datasets.