metadata
license: other
tags:
- eeg
- sequence-to-sequence
- in-context-learning
- brain-computer-interface
- large-eeg-model
library_name: pytorch
ECHO
ECHO: Toward Contextual Seq2Seq Paradigms in Large EEG Models is a decoder-centric framework for EEG modeling. Instead of treating a pretrained EEG encoder as a fixed feature extractor followed by a lightweight classifier, ECHO reformulates EEG modeling as sequence-to-sequence learning over EEG signals, task tokens, label tokens, and contextual support samples. This design is intended to improve cross-task and cross-dataset generalization and to enable in-context adaptation for heterogeneous EEG tasks.
Code
The official implementation is available at:
https://github.com/wythedee/ECHO
The public repository contains the encoder-decoder code structure:
FAST/: EEG encoder components.EEG2Text/: EEG-to-text decoder components.
Paper
- arXiv: https://arxiv.org/abs/2509.22556
- OpenReview: https://openreview.net/forum?id=ClLQ6cLkoR
@article{liu2025echo,
title={ECHO: Toward Contextual Seq2Seq Paradigms in Large EEG Models},
author={Liu, Chenyu and Deng, Yuqiu and Liu, Tianyu and Zhou, Jinan and Zhou, Xinliang and Jia, Ziyu and Ding, Yi},
journal={arXiv preprint arXiv:2509.22556},
year={2025}
}