IEEE Systems, Man, and Cybernetics Magazine - January 2018 - 39

0.9
0.8
0.7
0.5
0.4

Ps as a random variable and decomposes it into two factors
(Pdeploy and Peffectiveness) required for threat VT pairing. Each
factor is likewise treated as a random variable. We modeled
multiple VT pairs in series and in parallel to evaluate the
overall Ps . The AMADS then ran successive interactions to
conduct a Monte Carlo analysis that produced the level of
confidence achievable for selected VT pairs with respect to
their associated Ps estimations.
For the nonkinetic techniques, the results showed that
Pdeploy fell between 0.3 and 0.4 for the manufacturing and
fielding-and-deployment phases but dropped below 0.2 for
the boost phase. In contrast, Peffectiveness was below 0.2 for the
manufacturing and fielding-and-deployment phases
and fell between 0.3 and 0.4 for the boost phase. The
confidence intervals for these two factors of Pnegation were

Figure 7. Some objects of interest and their images

Small Color, Mainly
0.969
Land, No Object

Small Color, Mainly
0.975
Land, Object

Small Color, Mainly
0.608
Water, No Object

Small Color, Mainly
0.608
Water, Object

0.545

Large Grayscale,
SD, No Object

0.535

Peffectiveness

Pnegation

Large Grayscale,
No Object

Pdeploy

0.1

Large Grayscale,
SD, Object

0.2

an
M

0.6

0.3

Figure 6. The results for the missile-defense scenario.

and geolocations.

Ps for Each Path

Probability

Monte Carlo Probability (Averaged) Per Layer

u
(P fact
D
ha ur
ep
se ing
lo
0)
ym F
en ie
l
d
t ( in
Ph g
as an
e d
Bo
1)
os
t(
Ph
as
M
e
id
2)
co
ur
se
(P
ha
Te
se
rm
3)
in
al
(P
ha
se
4)

1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0

Figure 8. The remote-object-location scenario results.

SD: ship detection.

relatively tight. However, when computing Pnegation according to (5), the combined confidence factor was wide. These
results suggest that nonkinetic techniques with greater
effectiveness should be selected for the manufacturing
and fielding-and-deployment phases, and nonkinetic
approaches that can be deployed more successfully should
be used in the boost phase.
For the kinetic case, techniques for both the midcourse
and terminal phases yielded high Pk . However, the results
were not consistent across all of the AMADS runs; thus,
the standard deviation was large, and the corresponding
confidence intervals for both phases were wide. For these
examples, we seek kinetic techniques with more repeatable results.
Remote-Object-Location Scenario
Figure 7 shows a visualization screen for the remoteobject-location scenario results. It includes thumbnail
images of the objects of interest, along with their geolocations on a map. The circles on the map represent clusters
of images, and their colors represent the associated Ps and
confidence intervals.
The AMADS computed Ps, including confidence intervals, for each processed image along seven possible algorithm processing paths (i.e., six unique algorithms, with
one applied twice within the algorithm chain). To derive
confidence intervals, the DES ran Monte Carlo analyses in
which the input to each level was iterated many times to
derive a standard deviation for the Ps results. The actual results, shown in Figure 8, indicate that some paths
(i.e., 1, 6, and 7) yielded high Ps (above 0.9) with tight confidence intervals, while others (i.e., 2-5) provided lower Ps
(i.e., between 0.5 and 0.6) with wider confidence intervals.
Ja nua r y 2 01 8

IEEE SYSTEMS, MAN, & CYBERNETICS MAGAZINE

Table of Contents for the Digital Edition of IEEE Systems, Man, and Cybernetics Magazine - January 2018