Organizations: College of Computer Science, Nankai University, Tianjin 300350, China · State Key Laboratory of Multimodal Artificial Intelligence Systems, Institute of Automation, Chinese Academy of Sciences, Beijing 100190, China · Department of Neurosurgery, Beijing Tiantan Hospital, Capital Medical University, Beijing 100070, China · School of Computer Science and Cyber Engineering, Guangzhou University, Guangzhou 510006, China
Foundation models pre-trained on large unlabeled electroencephalography (EEG) corpora have emerged as a transformative paradigm for Brain-Computer Interface (BCI) applications. Recent EEG foundation models have shown that multi-scale representations derived from fixed-length patches can improve EEG representation learning. Existing frameworks commonly construct temporal scales through feature-level operations. However, these operations leave the correspondence between learned features and scale-specific characteristics of the original EEG signals implicit. Moreover, when the self-supervised objective is not aligned with the multi-scale input representations, information at some temporal scales may receive insufficient supervision during pre-training. In light of this, we propose ScaleBrain, a scale-consistent pre-training framework that explicitly coordinates multi-scale representations of EEG from both the model input and optimization objective sides. Specifically, ScaleBrain introduces multi-scale EEG tokenization to represent the same EEG segment through analysis windows with different temporal spans and shared temporal anchors, and pairs these representations with scale-specific reconstruction to align with the multi-scale input structure and provide supervision across temporal resolutions during self-supervised learning. We further develop an axial spatio-temporal attention mechanism to contextualize the fused multi-scale representation along spatial and temporal dimensions. Extensive evaluations on six public BCI benchmarks against 12 representative baselines show that ScaleBrain achieves the highest mean Balanced Accuracy on five datasets and ranks third on the remaining one, with only 5.0M parameters, demonstrating strong performance and favorable parameter efficiency.
Figures & tables
Fig. 1: Comparison of pre-training paradigms for EEG foundation models. Unlike (a) the existing multi-scale framework which focuses on the input side with different temporal anchors, our proposed (b) ScaleBrain explicitly maintains scale consistency by coordinating coarse-to-fine representations across input construction and self-supervised optimization. Long/mid/short-win: long/middle/short window.
Fig. 2: The overall framework of ScaleBrain, which is composed of (a) multi-scale EEG tokenization, (b) axial spatio-temporal encoder, and (c) scale-specific reconstruction. PE: Positional Encoding. FFN: Feed-Forward Network.
Task
Dataset
Rate (Hz)
Duration (s)
#Channels
#Samples
#Subjects
Label
Motor Imagery
PhysioNet-MI
160
4
64
9837
109
4-class
Emotion Recognition
SEED-V
1000
1
62
117744
16
5-class
Sleep Staging
HMC
256
30
4
137243
151
5-class
Mental Stress Detection
MentalArithmetic
500
5
20
1707
36
2-class
Abnormality Detection
TUAB
250
10
16
409455
2383
2-class
Event Type Classification
TUEV
250
5
16
112491
370
6-class
TABLE I: Overview of downstream BCI tasks and datasets.
Component
Setting
Input Channels
19
Time Points per Sample
6000
Mask Ratio
0.5
Mask Token
Learnable vector
Number of Layers
8
Hidden Dimension
192
TABLE II: Pre-training hyperparameters for ScaleBrain.
Component
Setting
Training Epochs
10
Batch Size
128
Optimizer
AdamW
Learning Rate
1e-4
Adam β
(0.9, 0.999)
Adam ϵ
1e-8
TABLE III: Default hyperparameter settings for downstream fine-tuning.
Method
Params
PhysioNet-MI (4-class)
SEED-V (5-class)
Balanced Acc.
Cohen’s Kappa
Weighted F1
Balanced Acc.
Cohen’s Kappa
Weighted F1
EEGNet
0.003M
0.5814 ± 0.0125
0.4468 ± 0.0199
0.5796 ± 0.0115
0.2961 ± 0.0102
0.1006 ± 0.0143
0.2749 ± 0.0098
EEGConformer
0.55M
0.6049 ± 0.0104
0.4736 ± 0.0171
0.6062 ± 0.0095
0.3537 ± 0.0112
0.1772 ± 0.0174
0.3487 ± 0.0136
SPaRCNet
0.76M
0.5932 ± 0.0152
0.4564 ± 0.0234
0.5937 ± 0.0147
0.2949 ± 0.0078
0.1121 ± 0.0139
0.2979 ± 0.0083
ContraWR
1.6M
0.5892 ± 0.0133
0.4527 ± 0.0248
0.5918 ± 0.0116
0.3546 ± 0.0105
0.1905 ± 0.0188
0.3544 ± 0.0121
FFCL
2.4M
0.5726 ± 0.0092
0.4323 ± 0.0182
0.5701 ± 0.0079
0.3641 ± 0.0092
0.2078 ± 0.0201
0.3645 ± 0.0132
TABLE IV: Comparison results of different methods across six EEG downstream tasks. The best , second-best , and third-best results are marked. * means recordings from TUAB and TUEV are excluded in pre-training. Acc.: Accuracy.
Setup
PhysioNet-MI
SEED-V
w/o Pre-training
0.6180 ± 0.0047
0.3957 ± 0.0048
w/o Scale-Aware PEs
0.6458 ± 0.0081
0.4111 ± 0.0088
w/o Multi-Scale
0.6358 ± 0.0079
0.4100 ± 0.0050
w/o MS Tokenization
0.6518 ± 0.0067
0.4054 ± 0.0095
w/o Scale-Specific Recon
0.6402 ± 0.0054
0.3969 ± 0.0052
Full
0.6524 ± 0.0111
0.4116 ± 0.0056
TABLE V: Ablation results reported in Balanced Accuracy (mean ± std). Note that the four scale-related setups form a 2×2 factorial decomposition over input-side and objective-side scale modeling. The best results are marked.
Fig. 3: Analysis of the necessity of scales. Acc.: Accuracy.
Fig. 4: Analysis of the effect of additional scales. Acc.: Accuracy.
Fig. 5: Analysis of the objective weights. Acc.: Accuracy.
Fig. 6: Analysis of the attention mechanism. Acc.: Accuracy.
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.
Recent electroencephalography (EEG) foundation models increasingly emphasize model scale, potentially limiting their deployment on resource-constrained devices and in everyday environments. We argue that transferable EEG features can instead be learned with a compact architecture and introduce BrainCodes, a lightweight foundation model pretrained through self-supervised learning on more than 800 hours of recordings from over 1,000 participants. Our architecture combines a quantization-based neural tokenizer with a spatiotemporal encoder that supports arbitrary electrode configurations. The pretraining objective explicitly encourages representations that capture spatial connectivity through channel positional encoding, and sensitivity to narrowband neural oscillations through an auxiliary training objective. Our model matches or surpasses the state of the art across diverse neurophysiological monitoring tasks, including sleep staging, attention assessment, and cognitive load estimation. BrainCodes transfers effectively to low-resource wearable devices while using up to 97% fewer trainable parameters than comparable architectures. Beyond predictive accuracy, our evaluation framework characterizes the geometry and invariances of learned EEG representations across measurable spatial, temporal, and spectral signal properties.
Kleanthis Avramidis, Ivan Tashev, Dimitra Emmanouilidou
Signal Analysis and Interpretation Lab, University of Southern California, Los Angeles, CA 90007, USA · Microsoft Research, Redmond, WA, USA
LaBraM, an early EEG foundation model, demonstrated that large-scale EEG pretraining can yield strong representations for diverse downstream tasks. Building on this foundation, we expand pretraining data and model capacity while retaining its core encoder and objective. We train ten encoders with 5.8 million to 1.97 billion parameters on three EEG collections spanning 25,067 to 79,910 hours, adapting signal preparation and training to heterogeneous recordings. Across 23 tasks, Atom-L* reaches 59.05% mean balanced accuracy at its best evaluated full fine-tuning (Full-FT) checkpoint, 3.61 percentage points above the strongest baseline EEG foundation model. Quark-G leads linear probing (LP) at 54.33%, a 1.24-point margin. Both also lead average task rank. At the last evaluated checkpoints, larger encoders improve Full-FT more consistently than LP, and stronger LP does not guarantee stronger Full-FT. Preserving patch structure improves LP over average pooling. Data-fraction experiments suggest capacity-dependent saturation, while checkpoint trajectories show that downstream performance can decline as pretraining loss falls. Joint classification with a shared Qwen3 readout reaches 50.50% with Molecule-H and Qwen3-8B; increasing language-model size provides no consistent gain. These results demonstrate the value of scaling LaBraM and the importance of matching pretraining scale to downstream adaptation.
Jun-Yu Pan, Ruicheng Yin, Bao-Liang Lu +3
Shanghai Jiao Tong University, Shanghai, China · Microsoft Research Asia, Shanghai, China · Fudan University, Shanghai, China