Scholarlink

This paper presents a new hybrid neural network for decoding linguistic representations from EEG signals. It consists of CNN layers and bidirectional LSTM networks with residual blocks in a four-stage progressive training approach using EEG recordings from Brennan and Hill during the story-listening dataset. Our model significantly improves classification accuracy across 628-word classes, greatly surpassing previous methods. Despite the variability among participants and the difficulty of generating synthetic data that mimics original EEG signals using GAN (Generative Adversarial Networks), this study offers a promising contribution to brain-computer interfaces with applications for individuals with communication disabilities.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Developing AI-Driven Techniques for Decoding and Visualizing Language Representations from Human Brain Activity

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Sahar Zidan Jleeb,
  • Dheyaa Shaheed Al-Azzawi

摘要

This paper presents a new hybrid neural network for decoding linguistic representations from EEG signals. It consists of CNN layers and bidirectional LSTM networks with residual blocks in a four-stage progressive training approach using EEG recordings from Brennan and Hill during the story-listening dataset. Our model significantly improves classification accuracy across 628-word classes, greatly surpassing previous methods. Despite the variability among participants and the difficulty of generating synthetic data that mimics original EEG signals using GAN (Generative Adversarial Networks), this study offers a promising contribution to brain-computer interfaces with applications for individuals with communication disabilities.