Biomedical Engineering Reference
In-Depth Information
1000
180
900
160
800
140
700
120
600
100
500
80
400
60
300
40
200
20
100
0
0
0
5
10
15
20
0
5
10
15
20
Iteration (j)
Iteration (j)
FIgURE 4.14:
Examples of the maximum absolute correlation curves over variable selection iterations
for (a) the original data and (b) the surrogate data.
From experiments with the surrogate data sets, the first threshold, δ
1
, is determined such that
few neuronal channels are selected in the surrogate data. The model parameters and the thresholds
are summarized in Table
4.6
. Note that δ
1
is determined much lower for more complex target hit-
ting data, indicating that the correlation of the FIR filter outputs with kinematics tends to be lower
for these data.
We show examples of neuronal subset selection results for the original and surrogate hand
reaching data, with the thresholds setting as described earlier, in Figure
4.15
. Figure
4.15a
shows
TaBlE 4.6:
The model parameters and neuronal subset selection thresholds
PaRaMETER
3d
2d
A learning rate for NLMS;
η
in (4.46)
0.1
0.3
A forgetting factor for FIR filter correlation;
μ
in (4.47)
0.95
0.95
A forgetting factor for channel covariance; ρ in (4.38)
0.8
0.9
A threshold for activation of variable selection; δ
1
0.7
0.3
A threshold for correlation in variable selection; δ
2
80
80
