Loading...
Search
| Friend's email | |
| Your name | |
| Your email | |
| enter code | |
This page was sent successfuly
Digital Object List
Bookmark
Review List
2 viewed
استخراج مولفه های سیگنال های مغزی با تجزیه تانسوری مبتنی بر یادگیری دیکشنری
سهرابی بناب، زهرا Sohrabi Bonab, Zahra
Event-Related Potentials Extraction using Dictionary-Based Tensor Decomposition
Sohrabi Bonab, Zahra | 2025
2
Viewed
- Type of Document: Ph.D. Dissertation
- Language: Farsi
- Document No: 58914 (05)
- University: Sharif University of Technology
- Department: Electrical Engineering
- Advisor(s): Shamsollahi, Mohammad Bagher
- Abstract:
- In electroencephalography (EEG) systems, brain activity is recorded noninvasively through an array of scalp electrodes, enabling the analysis of neural responses to external stimuli. However, interpreting these signals remains challenging due to their low signal-to-noise ratio (SNR), overlapping neural events, and substantial inter-trial and inter-subject variability. These difficulties stem from the inherent complexity, nonlinearity, and nonstationarity of neural data. This study is aimed at analyzing single-trial event-related potentials (ERPs) without relying on conventional averaging techniques, by adopting a structured modeling framework. The proposed approach is based on low-rank subspace recovery via tensor decomposition. Tensor factorization, as a powerful and flexible tool for modeling multi-dimensional data, enables the disentanglement of ERP signals into their fundamental spatial, temporal, and trial-wise components. Within this framework, by enforcing low-rank and sparsity constraints, the model suppresses noise and unstructured trial-to-trial variability, yielding more precise signal reconstructions. In addition to denoising, the method is employed for effective feature extraction, enhancement of classification accuracy, and subject-tosubject feature transfer. Empirical results demonstrate that by projecting the data onto a structured and shared latent subspace, the model reduces the need for extensive labeled data from new users, significantly decreasing system calibration time. To improve the interpretability of the decomposed components, dictionary learning is integrated as a complementary module. This integration preserves temporal structure in tensor factors and facilitates the clear identification of ERP components such as the P300, enhancing the transparency and explanatory power of the model. Moreover, comparative evaluations with baseline methods such as linear discriminant analysis (LDA) and advanced algorithms like Bayesian LDA (BLDA) reveal that the proposed model achieves competitive classification accuracy while substantially reducing runtime. Ultimately, by combining preprocessing, component extraction, and post-processing into a unified pipeline, this work introduces a comprehensive, interpretable, and computationally efficient framework for single-trial ERP analysis, offering a promising direction for real-time and practical brain-computer interface (BCI)
- Keywords:
- Electroencephalography ; Event Related Potential (ERP) ; Tensor Decomposition ; Low Rank Restoration ; Dictionary Learning ; Feature Transfer Learning
محتواي کتاب