Loading...

Detection of Event Related Potentials Using Tensor Decomposition

Jamshidi Idaji, Mina | 2015

2677 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 48772 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Shamsollahi, Mohammad Bagher
  7. Abstract:
  8. Tensors are valuable tools to represent EEG data. Tucker decomposition is the most used tensor decomposition in multidimensional discriminant analysis and tensor extension of LDA (Higher Order Discriminant Analysis-HODA) is a popular tensor discriminant method used for data of ERP-based BCIs. In this Thesis we introduce a new tensor-based feature reduction technique, named Higher Order Spectral Regression Discriminant Analysis (HOSRDA), with application in P300-based BCIs. The proposed method (HOSRDA) is a tensor extension of Spectral Regression Discriminant Analysis (SRDA) and casts the eigenproblem of HODA to a regression problem and therefore overcome the probable issue of singularity of scatter matrices. Also, the formulation of HOSRDA can provide the ability of adding regularization constrains to the factor matrices of the discriminant subspace. Additionally, when the dimension and number of data are very large, the regression problem can be solved via efficient iterative algorithms. We applied HOSRDA on data of a P300 speller from BCI competition III and reached average character detection accuracy 96.5%, 89%, and 72.5% for the two subjects, when 15, 10, and 5 repetitions are used for each intensification. HOSRDA has proved its high-accuracy performance in P300-based BCI and outperforms almost all the previous methods on the used dataset. HOSRDA, as a new multiway feature reduction technique, has a high performance for usage in P300-based BCI and can open a new framework for addingdifferent regularization constraints in higher order feature reduction problem. On the other side of the thesis, we have proposed the tensorized formulation of unsupervised classification in P300 speller and discussed the reasons of why it cannot perform at least the same quality of the vector-based formulation
  9. Keywords:
  10. Tensor Decomposition ; Tucker Decomposition ; Discriminant Analysis ; Electroencephalography ; Event Related Potential (ERP) ; Higher Order Spectral Regresion Discriminant Analysis (HOSRDA) Method ; Higher Order Discriminant Analysis (HODA) Method

 Digital Object List

 Bookmark

...see more