Upload
flavio
View
22
Download
0
Embed Size (px)
DESCRIPTION
DNA of DARE Dynamic Network Analysis Applied to Experiments from the Decision Architectures Research Environment. Kathleen M. Carley, Michael K. Martin, Carnegie Mellon University John P. Hancock ArtisTech , Inc. Challenge. - PowerPoint PPT Presentation
Citation preview
Advanced Decision Architectures Collaborative Technology Alliance
DNA of DAREDynamic Network Analysis
Applied to Experiments from the Decision Architectures Research Environment
Kathleen M. Carley, Michael K. Martin, Carnegie Mellon University
John P. HancockArtisTech, Inc.
Advanced Decision Architectures Collaborative Technology Alliance
Challenge
• Apply DNA to simulated battlefield data being generated in the DARE to produce tactically relevant insight
• Apply DNA techniques to complex problems where good solutions were lacking – Adversarial communications data – Distributed embedded Intelligent Agent messaging– Working toward real-time Monitoring and Prediction
• Support Army and DoD requirements in Intelligence and Net Centric Warfare problems
Advanced Decision Architectures Collaborative Technology Alliance
Two Case Studies
• Adversarial reasoning – intercepts of simulated communications
among humans• Movement in the perimeter
– Automated control of Persistent Coordinated Video Surveillance (PCVS) system
– simulated communications among software control agents & robots
Advanced Decision Architectures Collaborative Technology Alliance
Data Sources
ADA CTA Decision Architecture Research Environment (DARE)
• Persistent Coordinated Video Surveillance (PCVS) experiments
• Experimentally exploring the impact of automated reasoning on PCVS
AlgoLink Entity/Link Simulation• User specifies organizations (types,
sizes), locations, duration, frequency…
• Provides a “Ground Truth” file• Designed to test intelligence tools
Advanced Decision Architectures Collaborative Technology Alliance
DNA Tools
DyNetMLTable 1:
Know-ledge
Abdul Rahman Yasin
chemicals chemicals bomb, World Trade Center
Al Qaeda operative 26-Feb-93
Dying, Iraq palestinian Achille Lauro cruise ship hijackin
Baghdad 1985
2000Hisham AlHusse in
school phone, bomb
Manila, Zamboanga
second secretary
February 13, 2003,October 3,2002
Hamsiraji Ali
phone Abu Sayyaf, AlQaeda
Philippine leader
1980s
brother-in-law
Abu Sayyaf,Iraqis
Iraqi 1991
Name of Individual
Meta-matrix EntityAgent Resource Task-Event Organizati
onLocation Role Attribute
Abu Abbas Hussein masterminding
Green Berets
terrorist
Abu Madja phone Abu Sayyaf, AlQaeda
Philippine leader
Abdurajak Janjalani
Jamal Mohammad Khalifa, Osama binLaden
Hamsiraji Ali
Saddam Hussein
$20,000 Basilan commander
Muwafak al-Ani
business card
bomb Philippines, Manila
terrorists, dip lomat
Meta-Network
Unified Database(s)
Performance impact of removing top leader
64.5
65
65.5
66
66.5
67
67.5
68
68.5
69
1 18 35 52 69 86 103 120 137 154 171 188
Time
Perf
orm
ance
al-Qa'ida without leader
Hamas without leaderBuild Network -
Text Mining
Analyze –Statistics
SNA, DNA, Link
Analysis
Assess Change, What
if Analysis – Multi-agent
DNA
Advanced Decision Architectures Collaborative Technology Alliance
Case Study 1
• AlgoLink output delivered as XML file– Log of simulated comms intercepts– Each record identified sender, receiver, comm
time, comm duration, operational relevance of content, lat/lon of sender & receiver
• DNA strategy = overview + zoom– Engaged subset of *ORA capabilities– Geospatial visualization, key player ID, change
detection analysis, & correlation of standard and geospatial visualization.
– Did not use DNA text or simulation capabilities
Advanced Decision Architectures Collaborative Technology Alliance
Where is the Action?
Suspicious entities are fleeing Adelphi area over time course of scenario
Advanced Decision Architectures Collaborative Technology Alliance
How are they Organized?
FOG (Fuzzy Group Clustering) shows suspicious entities organized into 5
groups w/shared members.
════════ Interstitial members are likely to contain coordinators & leaders.
Advanced Decision Architectures Collaborative Technology Alliance
Who are the Key Players?
Drilling down…
*ORA’s Key Entity Report shows 3 agents critical to operations.
════════Narrow our focus from
set of interstitial members to small group of leaders.
Advanced Decision Architectures Collaborative Technology Alliance
When is the Action?
3 key players’ behavior changes at Period 3.════════
Agent 286 engaged in extensive coordination at Period 2. Reigns of control passed to Agent
652 at Period 4
A planning-execution phase-shift…════════
Organizational behavior changed in Period 2; radical difference by Period 3
Change Detection Analysis════════
Operation most likely in Period 3…Hidden, distributed structure coordinated into centrally controlled unit at
Period 3; Hiding again by Period 4
Advanced Decision Architectures Collaborative Technology Alliance
What Happened?
Period 3: Operation════════
Large cluster of suspicious entities in Adelphi area (with 286)
════════Cluster in apparent staging area (w/97)
════════Cluster associated with the runner, 652
Yellow = LocationRed = Agent
Bold Red = Key Player
Advanced Decision Architectures Collaborative Technology Alliance
Who was Where, When?
Period 4: Initial Surveillance════════
Key players never in same place at same time════════
Agents 286 & 97 (cyan & yellow) move about in their one region════════
Agent 652 (green) moving through many regions
*ORA Trails Viz════════
Time progresses down y-axis════════Geographic regions form
lanes on x-axis════════3 key players color-coded
Advanced Decision Architectures Collaborative Technology Alliance
Reasonable COAs
Period 6: End of Data Stream════════
COA 1: Scour Adelphi for bomb, IED, etc. planted during ops in Period 3════════
COA 2: Go after dispersed suspicious entities (286 may be an easy target, but the location where 97 is hiding will yield more suspicious entities)
Purple = LocationRed = Agent
Advanced Decision Architectures Collaborative Technology Alliance
Case Study 2: Scenario
ARTEMIS-PCVS System
════════ Tasking Agents control robotic surveillance assets in 1 of 4 quadrants to identify entities
moving within perimeter of Blue Force compound.
════════Scenario starts with short period of quiescence, followed by inject
of many moving targets that cross Tasking Agent AORs.
Advanced Decision Architectures Collaborative Technology Alliance
Case Study 2: Analysis
• Output from ARTEMIS-PCVS system delivered as XML file– Log of simulated communications among
software agents & robots– Each record identified sender, receiver,
communication time, message type• DNA strategy = converging operations
– More data but structurally redundant– Goal = detect rare handoff event where
Tasking Agents share robotic assets.
Advanced Decision Architectures Collaborative Technology Alliance
DNA Kick-start
• ArtisTech manually analyzed data– Meticulous message-trace analysis & event
identification– Identified message-types that indicate handoff– Identified number of handoffs
• CMU CASOS employed DNA techniques in *ORA to replicate ArtisTech’s analysis
Advanced Decision Architectures Collaborative Technology Alliance
Which Agents were Involved?
*ORA Sphere of Influence
════════Tasking Agent 3 shares.
Is positioned in network differently from others & sends/receives unique messages.
Advanced Decision Architectures Collaborative Technology Alliance
Psychological Validity of Newman Grouping
• Per ArtisTech, agents can be partitioned– Foreground agents substantive role– Background agents housekeepers
• Manual analysis took more than 4 hours• *ORA Newman Grouping in seconds
– 28 of 29 foreground agents correctly classified– 7 of 10 background agents correctly classified
• ArtisTech raters disagreed on status of 1 of the mismatched agents
Advanced Decision Architectures Collaborative Technology Alliance
Lessons Learned
• Open collaboration between data providers & network analysts creates beneficial gap between expected & observed multi-agent system behavior
• Dynamic Network Analytics foster– Understanding of emergent & reactive
behavior in multi-agent simulations (V&V)– Tactical insight and development of COAs
• DNA can be used to assess realism of data generation simulators
Advanced Decision Architectures Collaborative Technology Alliance
Next Steps
• Geo-spatial anchoring capabilities– Support shifting among perspectives provided by network, trails, and
map visualizations– Support locating key entities in different representations– Improve ability to correlate socio-network & geo-spatial viz
• i.e., automate generation of the annotated viz that shows where key players are in social network & on map
– Create capability for correlating trails viz & geo-spatial viz to give fine-grained spatio-temporal view of movement
• Tactical insight wizard– Codify set of analysis & viz techniques deemed useful for generating
COAs for different tactical situations– Reports would generate the entire DARE poster for example– Reports would differ in context-specific ways
• e.g., depending on whether the data are intercepts of communications of among humans (study 1) or software/robot agents (study 2)
Advanced Decision Architectures Collaborative Technology Alliance
Future
Time Frame• Low-hanging Fruit
• Intermediate Range
• Long Range
Product• Automate pre-processing for
*ORA input• Build & automate Tactical
Insight Report• DNA Monitoring of the
battlefield• Real-time DNA of evolving
battlefield• DNA based prediction