Paper Info
Title
The Effect of Limiting Trial Count in Context Aware BCIs: A Case Study with Language Model Assisted Spelling
Abstract
Deflections in recorded electroencephalography (EEG) in response to visual, auditory or tactile stimuli have been popularly employed in non-invasive EEG based brain computer intefaces (BCIs) for intent detection. For example, in an externally stimulated typing BCI, an accurate estimate of the user intent might require long EEG data collection before the system can make a decision with a desired confidence. Long decision period can lead to slow typing and hence the user frustration. Therefore, there is a trade-off between the accuracy of inference and the typing speed. In this manuscript, using Monte-Carlo simulations, we assess the speed and accuracy of a Language Model (LM) assisted noninvasive EEG based typing BCI, RSVPKeyboard (TM), as a function of the maximum number of repetitions of visual stimuli sequences and the intertrial interval (ITT) within the sequences. We show that the best typing performance with RSVPKeyboard (TM) can be obtained when ITI=150 ms and maximum number of allowed sequences is 8. Even though the probabilistic fusion of the language model with the EEG evidence for joint inference allows the RSVPKeyboard (TM) to perform auto-typing when the system is confident enough t o make decisions before collecting EEG evidence, our experimental results show that RSVPKeyboard (TM) does not benefit from auto-typing.
Year
DOI
Venue
2015
10.1007/978-3-319-20816-9_27
Lecture Notes in Artificial Intelligence
Field
DocType
Volume
Inference,Computer science,Brain–computer interface,Speech recognition,Spelling,Probabilistic logic,Language model,Electroencephalography,Visual perception,Rapid serial visual presentation
Conference
9183
ISSN
Citations 
PageRank 
0302-9743
1
0.38
References 
Authors
2
5
Name
Order
Citations
PageRank
Mohammad Moghadamfalahi184.26
Paula Gonzalez-Navarro211.05
Murat Akçakaya35921.15
Umut Orhan4608.66
Deniz Erdogmus51299169.92