Biomedical Engineering Reference
In-Depth Information
Figure 8.6
Computation of the Euclidian pattern classifier. For each time bin of a given spike
train the stimulus vector preceding this bin is assigned to one of two ensembles
(
P
(
|
=
)
(
|
=
)
) depending on whether the time bin contains a spike
or not. The Euclidian classifier is defined as the mean stimulus preceding spikes
(
m
1
, right) minus the mean stimulus preceding time bins without a spike (
m
0
,left):
f
s
r
0
and
P
s
r
1
=
m
1
−
m
0
. For this E-unit, the feature is a strong upstroke in amplitude, peaks at
around
25 ms, and then returns to 0 mV. Bandwidth of the stimulus: 0 to 44 Hz.
Adapted from [86].
−
can be conceived of as a measure of similarity between a stimulus segment and the
feature vector.
The performance of this Euclidian classifier in predicting the occurrence of spikes
is quantified using a Receiver Operating Characteristic (ROC) analysis [38, 40, 44,
86]. First, the conditional probability distributions of the projections,
P
(
h
f
,
(
s
)
|
r
=
1
)
q
and
P
(
h
f
,
(
s
)
|
r
=
0
)
,
q
are plotted and compared to threshold,
q
(
Figure 8.7b)
.
A spike is detected if
h
f
,
is larger than the threshold (to the right of the dashed
vertical line in Figure 8.7b). Integrating the tail of the distribution
P
(
s
)
>
0, that is if
f
;
s
q
(
h
f
,
(
s
)
|
r
=
1
)
q
Search WWH ::
Custom Search