Information Technology Reference
In-Depth Information
Object
Actions .views
.layers
View: Object
Actions Selections
Unit Text
VALUES
A
uto
Range
Display
X:
−−−
Y: −−− Z: −−−
Select
View
Move
ReShape
ReScale
Rotate
Receiving Unit
1.000
Build All
Connect All
0
Update
Init
New Layer(s)
New Label
Remove
Edit
0
0
0.95
0
0
0
0.95
0
0.95
0
Monitor Values
0.95
0
0
0
0.95
targ
ext
0.95
0
0
0
0.95
act
net
0.95
0
0
0
0.95
bias.wt
bias.pdw
0
0.95
0
0.95
0
act_eq
act_avg
0
0
0.95
0
0
act_m
act_p
Input
act_dif
da
vcb.hyst
vcb.acc
−1.000
Figure 2.18:
The NetView display of the detector network.
different digits, and these input units are connected to
the receiving unit via a set of weighted synaptic con-
nections (figure 2.18). Be sure you are familiar with
the operation of the NetView, which is explained in ap-
pendix A.8. We can view the pattern of weights (synap-
tic strengths) that this receiving unit has from the input,
which should give us an idea about what this unit will
detect.
inputs that will be presented to the network.
Press
View
on
the
main
control
panel
(
detect_ctrl
) and select
EVENTS
.
The window that comes to the foreground (an
En-
viroView
window, titled
Environment_0
)showsall
of the different
Events
in the
Environment
that will be
presented to a unit to measure its detection responses
(figure 2.19). The patterns that are presented to units
(and later to networks of units) are contained in an
envi-
ronment
because it is the environment in the real world
that provides the external inputs to brains. Each indi-
vidual pattern is contained in an
event,
which represents
one distinct possible state in the environment. As you
can see, the environment in this case contains the dig-
its from 0 to 9, represented in a simple font on a
5x7
grid of pixels (picture elements). Each pixel in a given
event (digit) will drive the corresponding input unit in
the network.
To see the unit respond to these input patterns, we
will present them one-by-one, and determine why the
unit responds as it does given its weights. Thus, we
need to view the activations again in the network win-
dow.
Press
the
r.wt
button on the lower left-hand side
of the NetView window (you will probably have to use
the scroll bar to find it — it's near the bottom of the list),
and then click on the receiving unit.
You should now see the input grid lit up in the pat-
tern of an 8. This is the weight pattern for the receiv-
ing unit for connections from the input units, with the
weight value displayed in the corresponding sending
(input) unit. Thus, when the input units have an acti-
vation pattern that matches this weight pattern, the re-
ceiving unit will be maximally activated. Input patterns
that are close to the target '8' input will produce graded
activations as a function of how close they are. Thus,
this pattern of weights determines what the unit detects,
as we will see. First, we will examine the patterns of
Search WWH ::
Custom Search