12. Brain-computer interface
? Aims to “decode” thoughts or commands from
human brain signal [Wolpaw+ 2002]
Encoding
Thoughts
Commands
Decoding Signal Acquisition
(EEG, MEG, …)
14. P300 speller system
A B C D E F
G H I J K L
M N O P Q R
S T U V W X
Y Z 1 2 3 4 ER detected!
5 6 7 8 9 _
A B C D E F
G H I J K L
M N O P Q R
S T U V W X
Y Z 1 2 3 4 ER detected!
5 6 7 8 9 _
The character must be “P”
15. 判別モデル
? 訓練サンプル (X1, y1), … , (Xn,yn)
X1 X2 X3 X4 Xn
– Xi は行列 (センサーの数 x 時間点)
– yi = +1 or -1 (2値分類)
訓練誤差 正則化
ただし (判別器)
18. Modeling P300 speller (decoding)
? Suppose that we have a detector f(X) that detects
the P300 response in signal X.
f1 f2 f3 f4 f5 f6
f7
f8
f9
f10
This is nothing but learning 2 x 6-class classifier
f11
f12
19. How we do this
… 12 2 8 1 3 4 11 9 5 6 10 7 …
Multinomial likelihood f. Multinomial likelihood f.
20. Experiment
? Two subjects (A&B) from BCI competition III
– 64 channels x 37 time-points (600ms @ 60Hz)
– 12 epochs x 15 repetitions x 85 letters = 15300 epochs in
training set
– 100 letters for test
? Linear detector function (bias is irrelevant)
36. 行列補完における相転移
観測pに対する行列の自由度 r(2n-r) の割合
全要素数n2に対する観測pの割合
Recht et al (2007) “Guaranteed Minimum-Rank Solutions of Linear Matrix Equations via Nuclear Norm Minimization”