Tuesday, April 5, 2011

Paper Reading #19: A POMDP Approach to P300-Based Brain-Computer Interfaces

Comments: Joe Cabrera, Patrick Firth.
Reference Information:
Title: A POMDP Approach to P300-Based Brain-Computer Interfaces
Authors: Jaeyoung Park,  Kee-Eung Kim, Sungho Jo.
Venue: IUI’10, February 7–10, 2010, Hong Kong, China.

Summary: This paper discusses mind reading using non-invasive EEG devices. In particular, it explores the area of text input. In traditional EEG mind reading text input, a grid of random characters is displayed. One by one, a character is flashed at random. If the user of the system were looking at the character as it was flashed, the EEG would pick it up, and the system would know which character they were thinking about.

The problem with this solution is that the algorithm is completely random. If you're trying to spell out the word "house", the flashing doesn't factor this in at all. It's just as likely to flash the letter J, Q, or Z at you, even though there's a good likelihood that you'd be attempting to input an S after the U.

The algorithm that the researchers have designed takes these sorts of things into account, as well as other particularities of the system, such as "Repetition Blindness". Repetition Blindness is a particularity of the system which takes place when the same letter is flashed immediately after or within close succession of itself. When this occurs, the system cannot differentiate between the letters nearby, and it must continue to flash randomly to find out which letter the user meant to input. The algorithm also takes this into account, and doesn't flash the same letter near itself.

Discussion: This article was incredibly interesting... but so technical. This isn't technically brain surgery, but it might as well be close. The overall gist of the article was interesting, but it's a little over the head of the average CHI journal reader. Overall though, it's fairly interesting to see someone implement cell phone style predictive text on EEG machines like this.

2 comments:

  1. This sounds really strange, I wonder how easy it is to use? While I don't quite understand this system, I agree, it does seem like it would be useful to implement predictive text for this application.

    ReplyDelete
  2. It's pretty amazing that the EEG device can tell which letter you're looking at by checking which part of the brain fires when the flash occurs.

    ReplyDelete