Would you like to add this to your personal library?
| Copyright © 2010. National Academy of Sciences. All rights reserved. Terms of Use and Privacy Statement |
Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter.
Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 389
Appendix C
Candidate DIME/PMESII Modeling
Paradigms
A
variety of modeling formalisms could be considered for DIME/
PMESII modeling efforts. We review some of them here.
Table C-1 compares selected modeling techniques by tabulat-
ing them against key characteristics which ultimately determine modeling
utility. In the remainder of this appendix, we define the characteristics, and
provide a very brief overview of each modeling formalism.
Expressivity of a modeling paradigm refers to its ability to capture
and express an analyst’s knowledge in terms of the constructs the para-
digm offers. The expressivity of a concept graph is very high as it keeps
the phrases used by the analysts intact in the model. In contrast, a neural
network model is only able to keep the input-output relationships in the
model. More expressive models are better able to capture the richness of
PMESII domains and are typically easier to build, use, and understand by
the modeler.
The executable feature of a modeling technique refers to whether some
useful information that is implicit in a model (e.g., degree of influence of
one variable onto another) can be derived from the model via some kind
of inferencing algorithm. A causal graph, for example, is an executable
paradigm as it offers propagation algorithms, and so also is a trained neural
network. In contrast, the concept-mapping model does not have such an
algorithm. Nonexecutable modeling techniques are useful for visualizing
complex models for human understanding and analysis; executable models
are useful for providing automated analysis of the models.
Reasoning of a modeling paradigm refers to the paradigm’s ability to
detect the direction of influence (not just connection) of one variable to
OCR for page 390
TABLE C-1 PMESII Modeling Paradigms and Their Characteristics
0
Characteristics:
Modeling Paradigms: Expressivity Executable Reasoning Adaptability Tools Exemplary Products
Concept map High No Forward Medium Free (Research) CmapTools (cmap.ihmc.us)
Backward
COTS Decision Explorer
(www.banxia.com/demain.html)
Concept graph Medium No Forward Low Free (Limited) Graphviz
(Graphviz) Backward (graphviz.org)
Yes GOTS OCCAM
(OCCAM) (http://www.cra.com/contract-r-d/
cognitive-systems-occam.asp)
Social networks Medium Yes Forward Medium GOTS OCCAM
Backward (cra.com/contract-r-d/
cognitive-systems-occam.asp)
Causal graph Medium Yes Forward Medium COTS BNet (cra.com/bnet)
Backward
Free (Limited) C4.5 (http://www.rulequest.com/
Personal/)
System dynamics Medium Yes Forward Low Free (Research) Ptolemy
model Backward (http://ptolemy.berkeley.edu)
Neural network Low Yes Forward High COTS NeuroSolutions
(www.neurosolutions.com)
Free (Research) Xerion
(www.cs.toronto.edu/~xerion/)
Situation theory Medium Yes Forward Low In-House PRISM (www.cra.com)
OCR for page 391
APPENDIX C
another. A belief network propagation algorithm, for example, incorporates
both deductive and abductive reasoning, and thus is able to detect both
forward and backward influences. On the other hand, the standard back
propagation neural-network modeling paradigm is limited only to forward
reasoning. Different modeling tasks require different kinds of reasoning.
It is sometimes useful to be able to look at a state and reason about likely
future outcomes (forward reasoning). For instance, one might want to
attempt to predict the likelihood of social unrest by evaluating the current
social, political, and economic state of affairs. Other times it is useful to
look at externally available information and diagnose the likely underlying
causes (backward reasoning). For instance, one might want to reason from
observed social unrest back to the likely underlying political, economic,
and social causes in order to properly address the causes of the unrest. For
these reasons, it is important to support both forms of reasoning with the
modeling tools we provide.
Adaptability of a modeling paradigm refers to automatic adjustments
by models, which are necessary to take into account new observations. It is
hard to adjust structures of graphical models as they are built in consulta-
tion with subject matter experts. But the strength of relationships among
a set of variables within a model (e.g., probabilities in a belief network
model or activation levels within a neural model) can be adjusted based
on observations without changing their structure. Having models that can
easily be adapted to represent new concepts and incorporate new data are
generally preferable.
Tools of a modeling paradigm refers to the currently available software
tools implementing the paradigm, that is, whether such a tool is commercial
off the shelf (COTS), government off the shelf (GOTS), open source, or
freely available for research/commercial purposes.
We now briefly describe the different modeling techniques shown in
the table.
CONCEPT MAPS
Concept maps are a result of research into human learning and knowl-
edge construction (Novak, 1998). In concept maps, the primary elements
of knowledge are concepts, and relationships between concepts are propo-
sitions. Concept maps are a graphical two-dimensional display of con-
cepts, connected by directed arcs encoding brief relationships (e.g., linking
phrases) between pairs of concepts forming propositions. Each concept
node is labeled with a noun, adjective, or short phrase, and each edge is
labeled with verbs or verb phrases describing the relation between the con-
nected concepts. Concepts maps are highly effective in quickly capturing
domain knowledge along DIME/PMESII dimensions.
OCR for page 392
BEHAVIORAL MODELING AND SIMULATION
A popular tool for concept mapping is the CmapTools (Canas et al.,
2004) package developed at the Institute for Human and Machine Cogni-
tion (see http://www.ihmc.us). The package is freely available for both
commercial and noncommercial use, and has many advantages over using
sticky notes or a more general diagramming tool (e.g., it can record the
entire mapmaking process). There are also COTS tools that can be used,
such as Banxia’s Knowledge Explorer.
CONCEPT gRAPHS
Concept graphs are a formal system of logic based on the existential
graphs of C.S. Peirce and semantic networks. Concept graphs explicitly
represent entities/concepts and relationships between entities as nodes in
a directed graph. They are mathematically precise and computationally
tractable structures, which have a graphic representation that is humanly
readable. For this reason, concept graphs have been used in a variety of
applications for computer linguistics, knowledge representation, informa-
tion retrieval, and database design. Their ease of use and generality make
them immediately useful for modeling a wide variety of domains, including
PMESII domains.
Figure C-1 is an example concept graph encoding a generic behavioral
model of a terrorist leader.
SOCIAL NETWORKS
Social networks are similar to concept graphs, but they represent social
structures. The nodes of the social network typically represent individuals
Terrorist Group A Leads Leader X Angers
Attr Attr Attr
Aggressive Diplomatic Quick to Anger
Causes Imminent Attack
Attr Attr
Attr
Use of Threatening Calling for Inviting Suicide
Phrases Jihad Bombers
FIguRE C-1 Concept graph model for terrorist leader behavior.
C-1.eps
OCR for page 393
APPENDIX C
and the links between them represent social relationships. Social network
analysis (SNA) provides tools for reasoning about social networks, their
strengths and weaknesses, the structural roles played by particular indi-
viduals, and their dynamics over time. Because of the focus on the analysis
of social structures, SNA is directly applicable to a range of PMESII model-
ing tasks.
SNA tools can be extended in a number of directions. For example,
one can build on traditional SNA functionality by providing additional
representational and analytic power by having nodes representing not only
individuals, but also arbitrary entities, especially including groups. Links
can be similarly extended to represent not only individual-to-individual
relationships, but also individual-to-group relationships (e.g., member-of)
and group-to-group relationships (e.g., rival political party). By providing
built-in Bayesian and rule-based reasoning capabilities, one could enable
automated analysis of the graph. For instance, a Bayesian network might
represent that members of a group might have a high probability of hold-
ing views that are promoted by that group, where the group, the indi-
vidual, and the ideology are all represented in the network as nodes with
appropriate links between them. In this case, an enhanced SNA tool could
automatically create a new believes link between the individual and the
ideology and annotate it with a particular probability.
CAuSAL gRAPHS
A causal graph (e.g., a belief network) (Jensen, 1996) is a graphical,
probabilistic knowledge representation of a collection of variables describ-
ing some domain. The strength of causal graphs are their ability to repre-
sent both the causal structure of a domain and the probabilistic elements of
those causal relationships (X causes Y with some probability), thus enabling
the modeling of both qualitative and quantitative details of the model.
In addition, the ability of causal graphs to handle both forward (causal)
reasoning and backward (diagnostic or abductive) reasoning makes them
especially well suited to domains with many sources of data, some of which
are uncertain, unreliable, or potentially missing. Many PMESII modeling
problems fall within such a scope.
Influence diagrams are a specialization of causal networks, augmented
with decision variables and utility functions to solve decision problems.
Decision trees are specialized influence diagrams that help to choose
between options by projecting likely outcomes as utilities. Such extensions
to causal graphs make it possible to also reason about the costs and ben-
efits of possible decisions. This functionality can be used to both support
intelligent decision making and to model likely decisions on the part of the
entities being modeled.
OCR for page 394
4 BEHAVIORAL MODELING AND SIMULATION
Bayesian reasoning tools, such as those provided by Microsoft (MSBN;
see http://research.microsoft.com/research/dtg/#bayesian), Norsys (Netica;
see http://www.norsys.com/index.html), and Charles River Analytics (BNet;
see http://www.cra.com), can support construction and reasoning with
causal graphs. There are also other existing COTS solutions to model-
ing influence diagrams and decision trees, such as C4.5 (see http://www.
rulequest.com/Personal/).
SySTEM DyNAMICS MODELS
As described in Chapter 4, system dynamics models, such as the
Stabilization and Reconstruction Operations Model (SROM) (Robbins et
al., 2005) can be used to analyze the organizational hierarchy, dependen-
cies, interdependencies, exogenous drivers, strengths, and weaknesses of a
country’s PMESII systems to enable more efficient resource expenditure.
SROM models PMESII systems at the national and regional levels, includ-
ing the interactions between regions. They also take into account demo-
graphic data, insurgent and coalition military, critical infrastructure, law
enforcement, indigenous security institutions, and public opinion.
The SROM models developed by the AFRL/IF NO’EM group were built
using the Ptolemy heterogeneous modeling software (see http://ptolemy.
berkeley.edu), which is developed and supported by the Electrical Engi-
neering and Computer Science department of the University of California,
Berkeley. While developed primarily for modeling of real-time embedded
systems, its heterogeneous processing model makes it an effective tool for
integrating a variety of data processing algorithms.
NEuRAL NETWORKS
A neural network is a nonlinear information-processing paradigm that
models complex systems with a large number of highly interconnected
processing elements (a.k.a. neurons or nodes), arranged in multiple layers,
working in unison to solve specific problems. Neural networks offer some
of the most versatile ways of mapping or classifying a nonlinear process
or relationship. Neural networks have been successfully used in diverse
paradigms, such as recognition of speakers in communications, diagnosis
of hepatitis, recovery of telecommunications from faulty software, inter-
pretation of multimeaning Chinese words, undersea mine detection, texture
analysis, three-dimensional object recognition, hand-written word recogni-
tion, and facial recognition. Neural networks would be useful in building
PMESII models for those domains that have highly complex nonlinear
relationships between input and output variables.
OCR for page 395
5
APPENDIX C
A large number of neural network construction kits and runtime
engines exist, including the Xerion tool from the University of Toronto (see
http://www.cs.toronto.edu/~xerion/) and the NeuroSolutions tools from
NeuroSolutions (see http://www.nd.com/products/nsv3.htm).
SITuATION THEORy
Situation theory models information processing and flow, that is, how
an agent extracts information from the world and how it is subsequently
transferred between agents. Situation theory provides a paradigm for
describing the world, an ontology for representing it, and a suite of infer-
ences for reasoning about it. Situation theory is unique in that it places
situations alongside individuals, relations, and locations as first-class mem-
bers of its ontology. Situations provide partial descriptions of the world
in terms of the features individuated by some agent. They are defined in
terms of the relationships they support; that is, they represent relationships
between relationships. Situations provide a powerful representation of
complex events spread over both space and time and, therefore, serve as a
natural representation of a variety of PMESII models. Situation theory has
been applied to a variety of fields including natural language understanding
(Barwise and Perry, 1983), information visualization (Lewis, 1991), coop-
erative social interaction (Devlin and Rosenberg, 1991), and both Level 2
(Steinberg and Bowman, 2004) and Level 3 (Steinberg, 2005) data fusion.
REFERENCES
Barwise, J., and Perry, J. (1983). Situations and attitudes. Cambridge, MA: MIT Press.
Canas, A., Hill, G., Carff, R., Suri, N., Lott, J., Eskridge, T.C., Gómez, Arroyo, M., and
Carvajal, R. (2004). CmapTools: A knowledge modeling and sharing environment. In
Proceedings of First International Conference on Concept Mapping (pp. 125–133).
Pamplona, Spain: Universidad Pública de Navarra.
Devlin, K., and Rosenberg, D. (1991). Situation theory and cooperative action. In Proceedings
of the Third Conference on Situation Theory and Its Applications, Oiso, Japan.
Jensen, F.V. (1996). Bayesian networks basics. AISB Quarterly, 4, 9–22.
Lewis, C.M. (1991). Visualization and situations. In J. Barwise, J.M. Gawron, G. Plotkin,
and S. Tutiya (Eds.), Situation theory and its applications II (pp. 553–580). Stanford,
CA: CSLI.
Novak, J.D. (1998). Learning, creating, and using knowledge: Concept maps as facilitative
tools in schools and corporations. Mahwah, NJ: Lawrence Erlbaum Associates.
Robbins, M., Deckro, R.F., and Wiley, V.D. (2005). Stabilization and reconstruction opera-
tions model (SROM). Presented at the Center for Multisource Information Fusion Fourth
Workshop on Critical Issues in Information Fusion: The Role of Higher Level Informa-
tion Fusion Systems Across the Services, University of Buffalo. Available: http://www.
infofusion.buffalo.edu/ [accessed Feb. 2008].
OCR for page 396
BEHAVIORAL MODELING AND SIMULATION
Steinberg, A. (2005). An approach to threat assessment. In Proceedings of the
Eighth International Conference on Information Fusion, Volume , Philadelphia, PA.
Available: http://ieeexplore.ieee.org/iel5/10604/33513/01592001.pdf?isNumber= [accessed
Feb. 2008].
Steinberg, A., and Bowman, C. (2004). Rethinking the JDL data fusion levels. In Proceedings
of National Symposium on Sensor and Data Fusion. Available: http://www.infofusion.
buffalo.edu/tm/Dr.Llinas’stuff/Rethinking%20JDL%20Data%20Fusion%20Levels_
BowmanSteinberg.pdf [accessed Feb. 2008].
Items in cart [0]
