Abstract:Electroencephalography (EEG) is a vital tool to measure and record brain activity in neuroscience and clinical applications, yet its potential is constrained by signal heterogeneity, low signal-to-noise ratios, and limited labeled datasets. In this paper, we propose FoME (Foundation Model for EEG), a novel approach using adaptive temporal-lateral attention scaling to address above-mentioned challenges. FoME is pre-trained on a diverse 1.7TB dataset of scalp and intracranial EEG recordings, comprising 745M parameters trained for 1,096k steps. Our model introduces two key innovations: a time-frequency fusion embedding technique and an adaptive time-lateral attention scaling (ATLAS) mechanism. These components synergistically capture complex temporal and spectral EEG dynamics, enabling FoME to adapt to varying patterns across diverse data streams and facilitate robust multi-channel modeling. Evaluations across four downstream tasks demonstrate FoME's superior performance in classification and forecasting applications, consistently achieving state-of-the-art results. To conclude, FoME establishes a new paradigm for EEG analysis, offering a versatile foundation that advances brain-computer interfaces, clinical diagnostics, and cognitive research across neuroscience and related fields. Our code will be available at <a class="link-external link-https" href="https://github.com/1061413241/FoME" rel="external noopener nofollow">this https URL</a>.

What problem does this paper attempt to address?

The problems that this paper attempts to solve are several key challenges in electrophysiological signals (EEG) in neuroscience and clinical applications: 1. **Signal heterogeneity**: EEG signals have a wide range of sources, and the acquisition systems used in different studies are different, resulting in significant differences in sampling rates, electrode positions, and numbers. Even following the international 10 - 20 system, it is still very difficult to strictly standardize the EEG acquisition protocol. 2. **Low signal - to - noise ratio**: EEG signals usually have a low signal - to - noise ratio, which complicates the effective extraction of useful information. 3. **Limited labeled data sets**: In the existing EEG data sets, the amount of labeled data is small, which limits the effectiveness of model training and makes large - scale signal annotation economically unfeasible. 4. **Insufficient model generalization and transfer ability**: Due to the inherent heterogeneity and low signal - to - noise ratio of EEG signals, existing models are difficult to perform effective transfer learning between different tasks. To solve the above problems, the author proposes a new method named FoME (Foundation Model for EEG), which utilizes self - supervised pre - training and an Adaptive Temporal - Lateral Attention Scaling (ATLAS) mechanism to deal with the complexity and diversity of EEG signals. Specifically, the main contributions of FoME include: - **Large - scale pre - training**: FoME has been pre - trained on a diverse data set containing 1.7TB of scalp and intracranial EEG recordings, covering more than 30,000 recordings with a total duration of approximately 26,000 hours. - **Time - frequency fusion embedding**: By integrating time - domain and frequency - domain features, a unified representation is generated, enabling the model to capture multi - scale time and spectral information. - **Adaptive Temporal - Lateral Attention Scaling (ATLAS) mechanism**: This mechanism can dynamically adjust the attention weights in the time and space dimensions, thereby more effectively capturing the changing patterns in different data streams and achieving robust multi - channel modeling. Through these innovations, FoME not only improves the performance of EEG signal classification and prediction tasks but also demonstrates its broad application potential in fields such as brain - machine interfaces, clinical diagnosis, and cognitive research. Experimental results show that FoME has reached the state - of - the - art level in multiple downstream tasks, verifying its effectiveness and versatility.

FoME: A Foundation Model for EEG using Adaptive Temporal-Lateral Attention Scaling

JDAT: Joint-Dimension-Aware Transformer with Strong Flexibility for EEG Emotion Recognition

Temporal attention fusion network with custom loss function for EEG--fNIRS classification

Graph-Enhanced EEG Foundation Model

EEGPT: Unleashing the Potential of EEG Generalist Foundation Model by Autoregressive Pre-training

EEGFormer: Towards Transferable and Interpretable Large-Scale EEG Foundation Model

ECG-FM: An Open Electrocardiogram Foundation Model

An Electrocardiogram Foundation Model Built on over 10 Million Recordings with External Evaluation across Multiple Domains

Neuro-GPT: Towards A Foundation Model for EEG

Large Brain Model for Learning Generic Representations with Tremendous EEG Data in BCI

EEGformer: A transformer-based brain activity classification method using EEG signal

Simultaneous EEG-fNIRS Data Classification Through Selective Channel Representation and Spectrogram Imaging

EEG-EMG FAConformer: Frequency Aware Conv-Transformer for the fusion of EEG and EMG

FetchEEG: a hybrid approach combining feature extraction and temporal-channel joint attention for EEG-based emotion classification

Automatic diagnostics of electroencephalography pathology based on multi-domain feature fusion

EEGFuseNet: Hybrid Unsupervised Deep Feature Characterization and Fusion for High-Dimensional EEG with An Application to Emotion Recognition

Enhancing EEG Signal Generation through a Hybrid Approach Integrating Reinforcement Learning and Diffusion Models

Foundation Models in Electrocardiogram: A Review

PyHFO: lightweight deep learning-powered end-to-end high-frequency oscillations analysis application

EEG emotion recognition approach using multi-scale convolution and feature fusion

EvoFA: Evolvable Fast Adaptation for EEG Emotion Recognition