跳转至

文章背景与核心概要

脑电图（EEG）解码模型由于采集协议和个体神经生理学差异带来的域偏移，往往在跨数据集和跨被试时泛化能力较差。为了克服这些挑战，本文介绍了 EEG-PRIME，这是一个专为跨数据集、多任务脑电解码而设计的新型两阶段基础模型。

该架构的核心创新包括：结合了频率截断频谱增强的掩码预训练、融合任务语义与特定数据集/被试不变性的指令微调、通过多级条件信号调节 Q-Former 的逐层查询调制机制，以及利用冻结的类别标签文本嵌入作为语义原型的预测方法。在涵盖运动想象、情感识别、ADHD 检测、内隐语音和心理工作负荷等 16 个数据集上的评估表明，EEG-PRIME 在跨被试设置下显著超越了现有的先进基线，并在未见数据集上展现出强大的零样本迁移能力。

EEG-PRIME: Prototype-Aligned Representation Learning with Multi-Level Conditioning for EEG Decoding

EEG-PRIME: Prototype-Aligned Representation Learning with Multi-Level Conditioning for EEG Decoding

Summary

Summary

Electroencephalography (EEG) decoding models frequently struggle with generalization across diverse datasets and subjects. This issue arises primarily from domain shifts caused by varying acquisition protocols and individual differences in neurophysiology.

Electroencephalography (EEG) decoding models frequently struggle with generalization across diverse datasets and subjects. This issue arises primarily from domain shifts caused by varying acquisition protocols and individual differences in neurophysiology.

To overcome these challenges, EEG-PRIME introduces a two-stage foundation model designed for cross-dataset, multi-task EEG decoding. The architecture features: 1. Masked Pretraining: An EEG encoder leverages masked reconstruction combined with frequency-cutoff spectral augmentation to learn robust, transferable representations. 2. Instruction Tuning: Incorporates task-semantic, dataset-specific, and subject-invariant conditioning. 3. Layer-wise Query Modulation: Modulates the Q-Former using multi-level conditioning signals. 4. Prototype-Aligned Prediction: Employs frozen text embeddings of class labels as semantic prototypes for cosine-similarity-based prediction across heterogeneous label spaces.

To overcome these challenges, EEG-PRIME introduces a two-stage foundation model designed for cross-dataset, multi-task EEG decoding. The architecture features: 1. Masked Pretraining: An EEG encoder leverages masked reconstruction combined with frequency-cutoff spectral augmentation to learn robust, transferable representations. 2. Instruction Tuning: Incorporates task-semantic, dataset-specific, and subject-invariant conditioning. 3. Layer-wise Query Modulation: Modulates the Q-Former using multi-level conditioning signals. 4. Prototype-Aligned Prediction: Employs frozen text embeddings of class labels as semantic prototypes for cosine-similarity-based prediction across heterogeneous label spaces.

Evaluated across 16 datasets encompassing motor imagery, emotion recognition, ADHD detection, covert speech, and mental workload, EEG-PRIME consistently outperforms state-of-the-art baselines under cross-subject settings. Furthermore, on two held-out datasets, it achieves competitive zero-shot transfer performance without requiring target-domain optimization, calibration, or linear probing.

Evaluated across 16 datasets encompassing motor imagery, emotion recognition, ADHD detection, covert speech, and mental workload, EEG-PRIME consistently outperforms state-of-the-art baselines under cross-subject settings. Furthermore, on two held-out datasets, it achieves competitive zero-shot transfer performance without requiring target-domain optimization, calibration, or linear probing.

Paper Metadata

Paper Metadata

  • arXiv ID: arXiv:2608.13072 [cs.AI]
  • Primary Subject: Artificial Intelligence (cs.AI)
  • Submission Date: August 13, 2026
  • Authors:
  • Shuailei Zhang
  • Muyun Jiang
  • Wei Zhang
  • Jinbo Chen
  • Zhiwei Guo
  • Yong Li
  • Yi Ding
  • Cuntai Guan
  • arXiv ID: arXiv:2608.13072 [cs.AI]
  • Primary Subject: Artificial Intelligence (cs.AI)
  • Submission Date: August 13, 2026
  • Authors:
  • Shuailei Zhang
  • Muyun Jiang
  • Wei Zhang
  • Jinbo Chen
  • Zhiwei Guo
  • Yong Li
  • Yi Ding
  • Cuntai Guan

Full-Text & Resource Links

Full-Text & Resource Links

Abstract

Abstract

脑电图（EEG）解码模型由于采集协议和个体神经生理学差异带来的域偏移，往往在跨数据集和跨被试时泛化能力较差。我们提出了 EEG-PRIME，这是一个用于跨数据集多任务解码的两阶段脑电图基础模型。EEG-PRIME 将掩码预训练与原型对齐的指令微调相结合，实现了跨不同 BCI 范式的指令感知和被试不变解码。在预训练期间，脑电编码器通过结合频率截断频谱增强的掩码重构来学习可迁移的表征。在指令微调期间，EEG-PRIME 引入了任务语义、特定数据集和被试不变的条件机制。由此产生的条件信号通过逐层查询调制（Layer-wise Query Modulation）调节 Q-Former，而类别标签的冻结文本嵌入则作为原型，用于在异构标签空间中进行基于余弦相似度的预测。在涵盖运动想象、情感识别、ADHD 检测、内隐语音和心理工作负荷的十六个数据集上进行的实验表明，在跨被试设置下，该模型相比最先进的基线和以往的脑电图基础模型取得了持续的性能提升。在另外两个未见数据集上，EEG-PRIME 无需目标域优化、校准或线性探测，便可达到与会话内校准模型相当的平衡准确率，展现出有前景的零样本迁移能力。

Electroencephalography (EEG) decoding models often generalize poorly across datasets and subjects due to domain shifts in acquisition protocols and individual neurophysiology. We propose EEG-PRIME, a two-stage EEG foundation model for cross-dataset multi-task decoding. EEG-PRIME combines masked pretraining with prototype-aligned instruction tuning to enable instruction-aware and subject-invariant decoding across diverse BCI paradigms. During pretraining, an EEG encoder learns transferable representations through masked reconstruction with frequency-cutoff spectral augmentation. During instruction tuning, EEG-PRIME incorporates task-semantic, dataset-specific, and subject-invariant conditioning. The resulting conditioning signal modulates the Q-Former through Layer-wise Query Modulation, while frozen text embeddings of class labels serve as prototypes for cosine-similarity-based prediction across heterogeneous label spaces. Experiments on sixteen datasets covering motor imagery, emotion recognition, ADHD detection, covert speech, and mental workload show consistent improvements over state-of-the-art baselines and prior EEG foundation models under cross-subject settings. On two additional held-out datasets, EEG-PRIME achieves balanced accuracy comparable to within-session calibration models without target-domain optimization, calibration, or linear probing, demonstrating promising zero-shot transfer capability.