PAINT Research Team Presentation Template

1
Does PNACO make airborne toxins? Whats the best
way to find out?
Unclassified/For Official Use Only
2
RAPIDRepresentation and Analysis
ofProbabilistic Intelligence Data
PAINT

• Carnegie Mellon University
• PI Prof. Jaime G. Carbonell / jgc_at_cs.cmu.edu /
  [phone]
• Dr. Eugene Fink / e.fink_at_cs.cmu.edu / (412)
  268-6593
• Dr. Anatole Gershman / anatoleg_at_cs.cmu.edu /
  [phone]
• DYNAMiX Technologies
• POC Dr. Ganesh Mani / gmani_at_dynamixtechnologies.c
  om / [phone]
• Mr. Dwight Dietrich / ddietrich_at_dynamixtechnologie
  s.com / [phone]

Unclassified/For Official Use Only
3
Motivation

• To create innovative management methods and
  computational tools for representation and
  analysis of uncertainty in large-scale data

Unclassified/For Official Use Only
4
PAINT Architecture
Unclassified/For Official Use Only
5
RAPID Solution Features

• Automatic impact evaluation of uncertainty on
  analyst hypotheses and conclusions
• Recommends optimized plans for proactive
  intelligence gathering
• Targeted reduction of system uncertainty by
  directed collection of critical data
• Based on industry-standard platforms Excel and
  Oracle
• Specialized enhancements for managing uncertainty
  calculus
• Easy-to-use no programming required

Unclassified/For Official Use Only
6
RAPID Value Proposition

• Provide analysts with an uncertainty decision
  framework answering the following questions
• What conclusions can be drawn from the available
  uncertain data, and at what level of certainty?
• What missing data is critical for verifying
  specific hypotheses, increasing the certainty of
  current conclusions, refuting tentative
  conclusions, or drawing new conclusions?
• What are the tradeoffs between the value of
  specific missing data and the difficulty of
  obtaining it?
• What is the most efficient way of verifying a
  given hypothesis by collecting additional data?
• What factors can change in the future, what are
  possible implications of these changes, and which
  contingencies should be analyzed?

Unclassified/For Official Use Only
7
Intelligence Analysis Workflow
Filtering and processing of new intelligence
Analysisof key indicators
Development of intelligence-collection plans
Propagation of inferences
New intelligence
Intelligencecollection
Analysts
Unclassified/For Official Use Only
8
UncertaintyManagement Architecture
Approved plans forproactive data collection
Hypotheses,conclusions, and data-collection
plans
Analystinterface
Uncertain situation assessmentand
data-collection planning
Value-addedreasoning tools
Uncertainty calculus andproactive probe planning
Proactiveintelligencecollection
Excel extension for the analysis of uncertainty,
probes, and proactive data collection
Newintelligence
Scalable assessment ofuncertain intelligence
Generalintelligencecollection
Relational database of uncertaindata and
inference rules
External API
PAINT SUB-SYSTEMS
Unclassified/For Official Use Only
9
Functional Components

• Uncertainty calculus and probe planning Analysis
  of incomplete data, identification of critical
  uncertainties, and selection of probes these
  tools are integrated into Microsoft Excel.
• Scalable assessment of uncertain intelligence A
  database of incomplete and uncertain facts,
  uncertain inference rules, and hypotheses, which
  allows scalable planning of proactive data
  collection these tools are integrated into
  Oracle.
• Analyst interface Integrated access to all
  system components, which extends the standard
  Excel interface.
• Value-added reasoning tools Optional tools for
  the automation of related reasoning tasks, such
  as entity co-reference, contingency analysis, and
  adversarial search.

Unclassified/For Official Use Only
10
Integration Features

• Novel uncertainty analysis tools built on
  industry-standard Excel and Oracle platforms
• Leverages all existing integration features of
  Excel and Oracle
• Familiar user interface allows users of Excel to
  quickly and easily build task-specific
  uncertainty analysis spreadsheets
• General-purpose Application Programming Interface
  (API) enables integrations with 3rd party
  software packages including other PAINT
  subsystems and components

Unclassified/For Official Use Only
11
Uncertainty Calculusand Proactive Probe Planning
Microsoft Excel
Unclassified/For Official Use Only
12
Generating Probes to Minimize Uncertainty
Proactive probe planning
Uncertainty analysis

• Representation of probes
• Evaluation of probe utility
• Automated selection andlaunching of critical
  probes

• Representation of probabilitydistributions and
  qualitativeuncertainty
• Uncertainty arithmetic

MICROSOFT EXCEL
Situation assessment
Contingency planning

• Representation of data utility
• Tracking utility changesduring data collection
• Identification of criticaluncertainties

What-if analysis of alternativefuture
developments and data-collection plans based on
anextension of Excel scenarios
Internal API
Uncertainty database
Value-addedreasoning tools
Processing ofdata streams
Unclassified/For Official Use Only
13
Novel Research (1 of 3)

• Management of uncertain data
• Semi-automated collection and analysis of
  uncertain data
• Enables analysts to create a database of
  uncertain knowledge
• Track changes in the available data and
  implications of these changes
• Underlying mechanisms include
• reasoning with uncertain data
• indexing structures for fast retrieval of exact
  and approximate matches
• inference rules for describing relations among
  data.

Unclassified/For Official Use Only
14
Novel Research (2 of 3)

• Hypothesis formulation and validation
• Semi-automated formulation of new hypotheses,
  their verification, and tracking changes in their
  estimated validity.
• Underlying algorithms include
• fast inference propagation
• game-theoretic analysis of goals
• possible actions of adversarial groups
• automated explanation of inferences to human
  analysts.

Unclassified/For Official Use Only
15
Novel Research (3 of 3)

• Data-collection planning
• Identification of critical missing data
• Analysis of its utility and costs
• Planning of proactive data-collection strategies
• We are developing
• active-learning techniques for the identification
  of key indicators
• game-theoretic techniques for the planning of
  proactive intelligence collection
• Analyst assistance for developing plans for
  resolving critical uncertainties at the lowest
  possible cost.
• mechanisms for the evaluation of trade-offs
  between the utility and costs of missing data,
  and contingency analysis of future developments.

Unclassified/For Official Use Only
16
Database Architecture
Uncertainfacts
Uncertaininferencerules
Semanticnetwork
Unclassified/For Official Use Only
17
Does PNACO make airborne toxins? Answer
Probably not, P ? 7
Unclassified/For Official Use Only