Scholarlink

Event-related potentials (ERPs), particularly the P300 component, hold significant importance across neuroscience, cognitive psychology, and brain–computer interfaces (BCIs). Detecting the P300 component is crucial for cognitive assessment, clinical diagnosis, and human–computer interaction. Nonetheless, due to the high subject variability and low signal noise ratio (SNR) of P300, its detection poses a persistent challenge. Deep learning (DL), notably convolutional neural network (CNN), presents advantages in P300 detection and has emerged as one of the foremost methods. However, compared to alternative methods such as linear discriminant analysis (LDA) and support vector machine (SVM), CNN’s complexity demands substantial resources for implementation. This chapter endeavors to develop a lightweight, efficient, and interpretable deep learning model for P300 detection, introducing the bantam-weight CNN (BCNN). As one of the simplest CNN architectures for P300 detection, BCNN consists of only one convolutional filter and a total of 165 parameters. Remarkably, it achieves state-of-the-art performance after just 2 epochs of training, establishing itself as an exceptionally lightweight and rapid CNN for P300 detection. Additionally, explainable AI (XAI) techniques are employed to elucidate its efficiency.

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

Toward Lightweight, Efficient, and Explainable Deep Learning Models for P300 Detection

  • 章节

  • 发表日期：

    2025年

查看全文
  • Maohua Liu,
  • Shi Wang,
  • Fred R. Beyette Jr.,
  • Liqiang Zhao

摘要

Event-related potentials (ERPs), particularly the P300 component, hold significant importance across neuroscience, cognitive psychology, and brain–computer interfaces (BCIs). Detecting the P300 component is crucial for cognitive assessment, clinical diagnosis, and human–computer interaction. Nonetheless, due to the high subject variability and low signal noise ratio (SNR) of P300, its detection poses a persistent challenge. Deep learning (DL), notably convolutional neural network (CNN), presents advantages in P300 detection and has emerged as one of the foremost methods. However, compared to alternative methods such as linear discriminant analysis (LDA) and support vector machine (SVM), CNN’s complexity demands substantial resources for implementation. This chapter endeavors to develop a lightweight, efficient, and interpretable deep learning model for P300 detection, introducing the bantam-weight CNN (BCNN). As one of the simplest CNN architectures for P300 detection, BCNN consists of only one convolutional filter and a total of 165 parameters. Remarkably, it achieves state-of-the-art performance after just 2 epochs of training, establishing itself as an exceptionally lightweight and rapid CNN for P300 detection. Additionally, explainable AI (XAI) techniques are employed to elucidate its efficiency.