Transcript Document

Domain Independent
Ontology Based Intelligence Vetting
using
Multiple Virtual Private Networks
and
Ontology Visualization
Dr. Paul Prueitt 1/29/2003
Producing Actionable Intelligence
Iterative Process Model
Human:
Understanding
Possible Outcomes
Measurement and
Instrumentation
Action-Perception cycle
Generating
Options
Persistent
Ontology Services
Representation and
Encoding
Reporting
Alerting
Detecting Facts
and Events
Producing and Matching
Models
Discovering
Relationships
Object Sciences Corporation 12/5/2002
Producing Actionable Intelligence
Technology Support
Presentation
to user
Analysis Tools and
Educational Processes
Visualization
Synchronous & Asynchronous
Collaboration Tools
Enterprise Middleware
Transaction Components
Heterogeneous
Databases
Object Sciences Corporation 12/5/2002
Data Schemas, Ontologies
Simple Reusable Architecture
Network of Virtual Private Networks
Data
Email and
Packet Transfer
HTTP
(SOAP,XML)
Applications
Connecter Architecture
Persistence
Object Sciences Corporation 12/5/2002
Collaboration
Visualization
Knowledge Sharing
Foundation
Education
Operational Architecture for Ontology Production
Data
Source
Schema-independent
data
Object Sciences Corporation 12/5/2002
Categorizer
and
Inference
Engines
Schema-dependent
data
Repository
Knowledge Sharing Foundation
Industry
Vendor Tools
KSF
Data
Core Engines
&
Educational Services
Repository
Real time
analysis
Micro-transaction accounting system
supporting outcome metrics and
revenue generation
Universities
Education
Distribution of revenue in compensation for use of tools or educational services
Important Innovation: Knowledge Sharing Foundation from The George Washington University
Diagram from Prueitt (2003)
Compensation for use of data, tools, educational
services, and work product
Data
Education
Vendor Tools
Embedded micro-transaction accounting system supporting
outcome metrics and revenue generation
Important Innovation: IP protected micro-transaction accounting system available from Dr. Brad Cox
Repository
Advanced data mining and natural
language processing
Making the case that new capabilities are within
reach
Differential Ontology Framework
By the expression “Differential Ontology” we choose to mean the interchange
of structural information between Implicit (machine-based) Ontology and
Explicit (machine-based) Ontology
• by Implicit Ontology we mean an attractor neural network system or one of the variations of
latent semantic indexing. These are continuum mathematics with only partial representation on
the computer.
• by Explicit Ontology we mean an bag of ordered triples { < a , r, b > }, where a and b are
locations and r is a relational type, organized into a graph structure, and perhaps accompanied
by first order predicate logic (such as the Topic Maps or Cyc ontologies). This is a discrete
formalism.
Explicit
Important Innovation: Differential Ontology from Dr. Paul Prueitt
Implicit
Diagram from Prueitt (2002)
Opponent-Ontologies based on
Latent Semantic Indexing
We use LSI in a specific fashion to produce a cognitive science
figure-ground corollary
C ground produces LSI transform,
T ground
Ground text collection,
C ground
T ground ( C exemplar )
forms the figure-from-ground
The figure-ground then measures the real time flow
of response passages
Exemplar text collection,
C exemplar
Measurement of Response I
We have identified three collections of semi-structured data (natural language)
Ground text collection, C ground , Response text collection, C response , Exemplar text collection, C exemplar
C exemplar
C ground
= B 1 B n B q
= { set of documents that are designed to cover linguistic variation that
needs to be picked up in a categorization process }
C response
= { the output from a web Harvesting system }
• C ground produces a linear algebra type transform
• This transform is defined from R
(vector) spaces).
m
into R
n
T ground .
(a transform between m and n dimensional Euclidean
• Exemplar sets, B i , I = 1, . . . , q, are made for each q categories of human-specified response.
• For example, response messages might be outputs from a harvester system that is measuring Arabic
response to world events.
Measurement of Response II
Finding the correct structure and content of the exemplar set is the key
C exemplar
T ground ( B i ) = N i
= B 1 B 2 B q
is the set of points (neighborhood) in R
n
formed by the i th exemplar bin
Creating these neighborhoods is easy. However, validating that the neighborhoods have
fidelity to the task at hand requires the use of an Ontology Lens that bring fidelity into
focus.
For text unit t e C response , T
ground
( t ) is a point in R n
The distance { T ( B i ) , T ( t ) } is now defined as the distance { centroid i , T ( t ) } where
the centroid is a point in R^300 the “stands in for” the image of the i th exemplar bin. The
“similarity” between a response passage and the linguistic contents of each of the exemplar
bins is approximated as a single positive real number.
The Ontology Lens shows structural relationships between the categories of the exemplar
set, and allows specialists to restructure the contents of the exemplar bins so that the
categories exhibit a high degree of independence, as measured by the degree of relationship.
Production of Concept Metrics
T ground (B 1)
Implicit
Ontology
T ground (B 2)
T ground (B 3)
d
Explicit
eC response
Ontology
T ground (B 5)
T ground (B 4)
For each response message, d, the implicit ontology produces a set of concept metrics, { mk },
and these concept metrics are used as the atoms of a logic. These atoms are used to produce an
explicit ontology. The logic is then equipped with a set of inference rules. Evaluations rules are
then added to produce an inference about opinions of the authors of the response set.
Important Innovation: Concept Metrics from Object Science Corporation
Diagram from Prueitt (2002)
Inter-Role Collaboration using Ontology
Knowledge Worker
Role and Event Specific View
Knowledge Repository
Synchronous
Collaboration
Periodic
Update
Knowledge Worker
Views
Object Sciences Corporation 12/5/2002
Tri-level Architecture
The Tri-level architecture is based on the
study of natural systems that exist as
transient stabilities far from equilibrium.
The most basic element of this study is the
Process Compartment Hypothesis (PCH)
that makes the observation that “systems”
come into being, have a stable period (of
autopoiesis) and then collapse.
Human cognition is modeled in exactly the
same way. Human mental events are
modeled as the aggregation of elements of
memory shaped by anticipation.
Important Innovation: Tri-level architecture from Dr. Paul Prueitt
The tri-level architecture for machine
intelligence is developed to reflect the
PCH. A set of basic event atoms are
developed through observation and human
analysis. Event structures are then
expressed using these atoms, and only
these atoms, and over time a theory of
event chemistry is developed and reified.
Diagram from Prueitt (1995)
cA/eC
Neuroscience informs us that the physical process that brings the human experience of the past
to the present moment involves three stages.
1) First, measured states of the world are parceled into substructural categories.
2) An accommodation process organizes substructural categories as a by-product of
learning.
3) Finally, the substructural elements are evoked by the properties of real time stimulus to
produce an emergent composition in which the memory is mixed with anticipation.
Each of these three processes involves the emergence of attractor points in physically distinct
organizational strata. The study of Stratified Complexity appeals first to foundational work in
quantum mechanics and then to disciplines such as cultural anthropology and social-biology.
categoricalAbstraction (cA) is the
measurement of the invariance of data
patterns using finite set of logical atoms
derived from the measurement.
eventChemistry (eC) is a theory of type
that depends on having anticipatory
processes modeled in the form of
aggregation rules, where the
aggregation is of the cA logical atoms.
Important Innovation: eventChemistry from Dr. Paul Prueitt
Diagram from Prueitt (1995)
gF/cA/eC
Evocative generalFramework (gF) theory constructs cA/eC knowledge bases directly in “conversation” with humans
We have projected a physical theory of structural
constraint imposed on any formative processes,
to a computational architecture based on
frameworks.
Various forms are conjectured to exist as part of
emergent classes, and in each case each class of
emergent types has a periodic table – like, in
many ways, the atomic period table.
The Sowa-Ballard Framework has 18 “semantic
primitives”.
Ballard/Sowa Framework
“According to Alvin Toffler, knowledge will become the central
resource of the advanced economy, and because it reduces the need
for other resources, its value will soar. (Alvin Toffler, Power Shift,
1990). Data warehousing concepts, supported by the technological
advances which led to the client/.server environment and by
architectural constructs such as the Zackman Framework, can
prepare organizations to tap their inner banks of knowledge to
improve their competitive positions in the twenty-first-century.
Zackman Framework
Important Innovation: Framework software from Drs. Paul Prueitt and Richard Ballard
Diagram from Prueitt (2001)
Situational Logic Construction
A latent technology transform, T ground , is used to produce
simple metrics on membership of documents from the
response collection C response in the categories defined by
the contents of the bins
C exemplar
= B 1 B n2 B q
These bins are represented in the situational logic as the logical atoms A , from which a
specific logic is constructed. These atoms are then endowed with a set of q real numbers that
are passed to an Inference Processor.
The set of q real numbers are computed from a formal “evaluations of the structural
relationship between logic atoms” using the Ontology Lens.
Atom a  { r1 , r2 , . . . , rq }
The process of developing a situational logic is to be modeled after quasi axiomatic theory.
In this model, new data structure are in-put as axioms, and then a process of reduction of
axioms to logical atoms occurs.
The reduction also requires the Ontology Lens, (invented 2002 by Prueitt).
Important Innovation: Situational logics from Paul Prueitt
Diagram from Prueitt (2002)
The Ontology Len (discovered by Prueitt, 2002) is a structural focus “instrument” that is
designed to allow non-computer scientists to specify high quality exemplar sets. This is
done with an Implicit Ontology to Explicit Ontology (IO-EO) loop.
• When the user puts a new unit into a bin or removes a unit from a bin, then the IO-EO loop
will produce a different result.
• It is the human responsibility to govern the IO-EO loops so that the results have the
properties that the human wants, mostly independence of categories, but perhaps some
specific (and maybe interesting) “category entanglements”.
A graphic representation of what we call a “LSI structural similarity
matrix”.
The similarity is called structural because the exact notion of
semantic similarity is not known from this algorithmic computation
by itself.
The paragraphs of a small exemplar set (see appendix A) are
ordered as labels for the columns and rows.
One would expect that a paragraph would be structurally similar
with itself, and this is in fact what one sees as a set of dots
(representing a value of 1)
down the diagonal.
Diagram from SAIC (2002)
Minimal Work Flow
Production of the Explicit Ontology
C response = C response = C response
Implicit T
Ontology
ground
(B 4)
T ground (B 5)
T ground (B 3)
T ground (B 2)
T ground (B 1)
Ontology
Lens
Schema-independent
data
Schema-independent data is developed from the Ontology Lens, in the form of a set of syntagmatic units
{ < a, r , b > }
Where a and b are categories defined by the exemplar set, , and r is a measure of relationship.
Searching and Filtering
Storing
Analyzing Entities
Visualizing Links
Clustering
Categorizing
Resolving Cover Terms
Matching Models / Detecting
Changes
Simulation
Generating Hypotheses
Generating Threat Scenarios
Structured Argumentation
Learning Patterns
Understanding Intent
Performing Risk Analysis
Generating Options
Generating Plausible Futures
Storytelling
Creating Explanations
Alerting
Visualizing GIS Data
Understanding Policies
Preparing Video Sources
Processing Text Sources
Processing Sensors
Processing Audio Sources
Translating Languages
Identifying Humans
Summarizing Data
Summarizing Text
Searching and Filtering
Categorizing
Indexing
Visualizing Summaries
Collaboration
Presenting Recommendations
Presenting Analysis Results
Presenting Situation Status
Presenting Options
Building Teams
Building Context
From the Intelligence
Community and DoD
Lessons Learned
To the Center for
Disease Control and
University Research
Centers
Synchronous & Asynchronous
Collaboration Tools
Visualization
Analysis Tools and
Educational Processes
Producing Actionable Intelligence
Iterative Process Model - decomposition of function/structure
V, S, R, K
Human:
Understanding
Possible Outcomes
V, N, S, R, K
Action-Perception cycle
V, S, R, K
Measurement and
Instrumentation
Generating
Options
V, S, R, K
V, D, N, S, R, K
Persistent
Ontology Services
Representation and
Encoding
Reporting
Alerting
S, R, K
Detecting Facts
and Events
V, D, N, S, R, K
V: visualization,
D: data mining,
Producing and Matching
Models
N: natural language processes,
S: support decision making,
V, D, N, S, R, K
R: structuring of reasoning
Discovering
Relationships
V, D, N, S, R, K
K: knowledge representation
T: technical support