Home > Publications > Doctoral dissertation & Master thesis
Domestic Conference
Title Classification improvement for a P300 speller brain-computer interface
Degree Ph.D.
Author Kyungae Yoon
Advisor Kiseon Kim
Graduation Date 2018.02.20 File Link icon
    Date 2018-03-28 21:17

 Classification improvement for a P300 speller brain-computer interface

A brain-computer interface (BCI) enables a direct communication pathway b etween a human brain and an external device. The purpose of the BCI is to provid e communication capabilities to people suffering from motor disabilities such as spinal cord injuries, amyotrophic

lateral sclerosis (ALS; i.e., Lou Gehrigs di sease). The BCI research has been accelerated and taken great strides toward mak ing

a BCI a practical reality to people suffering from motor disabilities. Never theless, the end goal has still not been reached

and there is much work to be do ne to produce real world worthy systems that can be comfortably, conveniently, a nd reliably used by

people with motor disabilities.


A BCI can be considered as a pattern recognition system that identifies patterns of brain activity. Underst anding a feature and a

classification algorithm is important to make a BCI a rel iable pattern recognition system. In order to design the most appropriate

classi fier for a given feature set, it is essential to clearly understand what feature s are used, what their properties are and how

they are applied.


In this thesis, we propose multiple kernel learning (MKL) based on three discriminant features to learn an reliable P300 classifier to

improve the accuracy of character recogn ition in a P300 speller BCI. A linear kernel is established for each discriminan t feature. A

kernel weight differentiates the linear kernel to both explore comp lementary information among the three discriminant features and

weigh a contribu tion of each discriminant feature for the MKL. The L1-norm regularization of the kernel weight ultimately

enforces an optimal discriminant feature set of the MK L of a P300 classification. Compared to an existing SVM-based classification

met hod, the proposed method consistently obtains better or similar accuracy in term s of character recognition for the variable size of

the three discriminant featu re sets.

광주과학기술원 한·러 MT-IT 융합기술연구센터 광주과학기술원정보통신공학부