Biomedical Engineering Reference
In-Depth Information
FIR HP
FIR
HP
RM
L
P H
P
1
1
PP
2
0.5
3
4
M1
0
5
0
50
100
RMLP HP
6
1
PMd
7
0.8
0.6
8
0.4
M1/
PMd
ipsi
9
0.2
10
0
0
50
100
Neuron Rank
FIgURE 4.6:
Reaching task neuronal sensitivities sorted from minimum to maximum for a movement.
The 10 highest sensitivities are labeled with the corresponding neuron.
effect of computing the sensitivities through both WF and RMLP topologies, we have found that
sensitivity-based selection of neurons is not heavily dependent on the model topology, even with
two distinct topologies (linear-feedforward vs. nonlinear-feedback) are utilized [
4
].
As a further check of the sensitivity importance measure, we trained three identical RMLP
networks of five hidden PEs, one with the 10 highest sensitivity neurons, one with the 84 inter-
mediate sensitivity neurons, and one with the 10 lowest sensitivity neurons. A plot of the network
outputs in Figure
4.7a
,
b
, and
c
shows that the reduced set of highest sensitivity neurons does a good
job of capturing the peaks of the movement, whereas the lowest sensitivity neurons are unable to
discover the mapping from spike trains to hand movements. The remaining intermediate sensitiv-
ity neurons contain information about the movement but it is not enough to capture the full detail
of the trajectory, although it does improve the overall trajectory fitting. Using the cumulative error
metric (CEM)
5
for BMI performance in movements, we see that the curves approach the bisector of
the space as sensitive neurons are dropped from the mapping, indicating a decrease in performance.
5
CEM(
r
) is the estimated probability that the radius of the error vector is less than or equal to a certain value
r
.
