The content discusses the challenges faced by deep learning models in fNIRS when identifying and excluding abnormal data. It introduces a method combining metric learning and supervision to improve model performance. The experiments show significant enhancements, particularly with transformer-based models.
The study highlights the importance of accurately classifying human behavioral intentions using fNIRS signals. Traditional methods like Linear Discriminant Analysis are compared to deep learning approaches like CNN and LSTM for feature extraction. The article emphasizes the need for reliable brain-computer interfaces supporting individuals with disabilities.
Furthermore, the content delves into the dataset used, preprocessing steps, network structures, training methodologies, and experimental results. It showcases how two-stage training can effectively exclude out-of-distribution data while maintaining classification accuracy. Visualization of feature vectors in detector and classifier subspaces provides insights into model performance.
Overall, the study suggests that integrating metric learning with supervised methods offers a promising solution for improving the reliability of deep learning models in fNIRS classification tasks.
翻譯成其他語言
從原文內容
arxiv.org
深入探究