Loading...
Please wait, while we are loading the content...
Similar Documents
Learning Priors for Error-related Decoding in EEG data for Brain-Computer Interfacing
| Content Provider | Semantic Scholar |
|---|---|
| Author | Schultheis, Matthias |
| Copyright Year | 2016 |
| Abstract | Brain-Computer Interfaces (BCIs) offer the possibility of controlling external devices by modulating brain activity. A user can send commands to control the BCI by performing mental tasks while wearing an electroencephalography (EEG) cap. EEG-based systems can already be used as a way for basic communication; however, due to nonstationarity of the brain signals, a system is only valid for a short amount of time after training. To provide remedy for this problem, an adaptive system can be built which adapts to changed brain signals and therefore stays suitable for a long period of time. In this thesis, an adaptive approach is presented introducing a prior over commands. In the decoding phase, this prior balances commands in the case where a traditional BCI becomes biased to specific commands. For applying such a system, brain signals need to be labeled online, i. e. the correct commands for control have to be known. Often, this information is only known by the user and unknown for the system. In the case of unknown labels, the system can infer feedback information by analyzing the user’s brain signals shortly after presenting visual feedback about decoded commands. Neurobiological research found error detection mechanisms in the human brain which can be used for decoding feedback. Feedback indicating incorrect decoding of the signals leads to particular time-bound deflections in EEG signals. These deflections can function as features to infer correct commands for establishing an adaptive BCI. This thesis investigates feedback decoding in the case of individuals playing a video game. This game is currently used for testing BCI systems for the use in daily tasks. In the game the user is supposed to send commands using an EEG system and receives feedback about the decoded commands. An experiment where subjects were supposed to play this game shows that the information whether any feedback was perceived can be decoded with an accuracy of 76.6% on average. In the presented experimental setup, the decoding whether correct or wrong commands were sent reaches an accuracy of 66.3%. For the discrimination analysis, subsampled signals of one electrode in the time domain are used as features and linear discriminant analysis and support vector machines are used as classification methods. Additionally, variations of the analysis such as using spatial and frequency-based features and additional preprocessing are considered. |
| File Format | PDF HTM / HTML |
| Alternate Webpage(s) | http://www.ausy.tu-darmstadt.de/uploads/Main/Abschlussarbeiten/Schultheis_BScThesis_2016.pdf |
| Language | English |
| Access Restriction | Open |
| Content Type | Text |
| Resource Type | Article |
