Upload
others
View
3
Download
0
Embed Size (px)
Citation preview
Towards a Comprehensive Computational Experimentation Environment: Dynamic Data
Driven Approach
Alok Chaturvedi, PIPurdue Homeland Security InstitutePurdue Homeland Security Institute
Purdue UniversityPurdue University
NSF ITR Grant # CNS-0325846, Indiana 21st Century R&T Fund, JPEOCBD, JFCOM J-9
Outline• Background• Building the Synthetic Environment
– Society of Simulations– Configuration– Integration
• Calibration• Validation• Experiment• Q&A
Support & Partnerships:
National Science FoundationIndiana State 21st Century Fund
Department of DefenseIndiana State Department of Health
Purdue University (PHSI, CRI, Discovery Park, Krannert School of
Management, School of Liberal Arts, College of Engineering, and Department of Computer Sciences)
Simulex Inc.
TEAM• PI: Alok Chaturvedi
• Co-PIs: Roko Aliprantis, Steve Beaudion, Jerry Busemeyer, Daniel Dolk, Jay Gore, Anant Grama, Chris Hoffman, Elias Houstis, Rich Linton, Shailendra Mehta, Suresh Mittal, Jenna Rickus, Vernon Rego, Ahmed Sameh, and Rusi Taleyarkhan.
• Collaborators: Eric Dietz, Jari Niemi, Chih-hui Hsieh, Teja Bhatt, Angela Mellema, Beth Naylor, Cliff Wojtalewicz, Rashmi Chaturvedi, Chee Foong, Midh Mulpuri, David Lengachar, Yeeling Tham, Steve Mallema, Johnson Char
• State of Indiana• Purdue University, West Lafayette, IN, USA • Naval Postgraduate School, Monterey, CA, USA• Indiana University, Bloomington, IN, USA• Simulex Inc.
Computational Experimentation Paradigm
WORLD
Observations
BehaviorEstimation
SyntheticEnvironment
Theory+DataValidation
Assumptions Model
Observations
Theories ofPhenomena
Validation
ConditionalPredictions
AssumptionsModel
SyntheticObservations
Theories ofPhenomena
Validation
ConditionalPredictions
Analytical &Observational
ComputationalExperimentationDomain Models
& Theories
Human
Issues in Computational Experimentation..
• High fidelity computational experimentation requires deep understanding of:
– The underlying Science -- physical, social, computational, life sciences, humanities
– The computational models • Mathematical equations• Equation free• Differential equations vs Difference equations• Baseball or basketball
– Representation paradigms -- Common (?) Uniform (?) Diverse (?)
– Presentations -- digital art (?) semantics (?) ontology (?) storytelling (?) entertaining (appeals to emotions)
– Platform HPC (?) Peer-2-Peer (Xbox, PS 2/3, PDA)– Business models (?)
Outline• Simulations Composed of Simulations• A Society of Simulations: the Concept
– Defining a Society of Simulations– The Components of a Society
• Fire Evacuation Society: a Case Study– The Goal– The Heterogeneous Members– Bridging the Members– Resulting Storylines
• Future Work– Further Development & Lessons Learned– Other Applications
Society of SimulationsA Definition
society:1. A group of human beings bound together by shared institutions and
culture.2. An association of people with mutual aims or interests.
• Each simulation is an autonomous member of the society.• Members operate independently on their own understanding of
reality.• Members share overlapping portions of their models of reality.• Members of a society cooperate to achieve their goals.• The fulfillment of the societal goals emerge from member actions.
Society of SimulationsExample: Fire Evacuation Society
• Goal:Model evacuation and health issues to address safety concerns regarding building fires.
• Members: Fire simulation & synthetic human simulations.
Fire Sim.
Human Sim.Human Sim.Human Sim.
building,smoke,temperature
A Society of SimulationsShared Reality
Portions of reality modeled by multiple members.• A shared and distributable data space.• Data is pulled by the members
on demand.• Decouples publishing & subscribing.
All interactions occur within the shared reality.• Time coordination by waiting for unavailable data.• Allows for multiple spatial & temporal granularities of access.
CD
BA
sharedreality
building,smoke,temperature
Fire Sim.
Human Sim.Human Sim.Human Sim.
A Society of SimulationsThe Bridging Component
Links between members and shared reality.• Grabs the data from shared reality that influences its member &
writes data the member produces.• Member-specific ontology translation.• Translates, scales, filters, & interpolates data.• Can trigger a series of actions in the member.• (optimistic) Initiates a roll back when significant data in the
member’s past is available.
Fire Sim.
Human Sim.Human Sim.Human Sim.building,
smoke,temperature
A Society of SimulationsThe Member Component
Autonomous & specialized models.• Developed independently.• Can be event-based or continuous-time,
optimistic or conservative.• Reusable in multiple societies.• Sense shared data (on-demand.)• Interpret data relative to member-specific semantics.• All interactions with members is through shared reality.
Fire Sim.
Human Sim.Human Sim.Human Sim.building,
smoke,temperature
Shared Reality Model
Shared Reality Engine
Fire Evacuation SocietyMembers in General
Fire Simulation:• FDS (Fire Dynamics Simulator,
from NIST.)• Optimistic
(checkpoint / restart.)• Continuous-time,
fluid dynamics.• Fine-grained in space
and time.• Time-consuming
(1 minute required to simulate each second.)
Human Simulation:• Includes physiology, mobility,
familial relationships, & some decision-making.
• Discrete-event.• Coarse-grained.• Real-time.• Conservative
(follows a leader.)
Fire Evacuation Society
• Children playing with matches.
• Kitchen fires.
Integrated Science Based Modeling
Bridged Science and Agent based models• Fire Model (NIST’s FDS)• Structure Model (LS Dyna)• Agent based Model (SEAS)• Physiology Model
• Safer building• Virtual Ground Zero• Prognosis for Real-time courses of
action planning
Large Evacuation Model
SEAS Parallel World Architecture
JINI Network Technology Services
RMID Services
Discovery Services
Transaction Services
Output Space
Output Space
Agent Space
Agent Space
Input Space
Input Space
Parallel World 3
Worker Agents
Proxy Service
Output Space
Output Space
Agent Space
Agent Space
Input Space
Input Space
Parallel World 2
Worker Agents
Proxy Service
Output Space
Output Space
Agent Space
Agent Space
Input Space
Input Space
Parallel World 1
Worker Agents
Proxy Service
Parallel World Coordination Space
Parallel World Coordination Space
Action Permutation
Algorithm
Exit Effect 1
• All 5 exits open: 4 deaths
Exit Effect 2• Top exit closed: 18 deaths
Exit Effect 3
• Bottom exit closed: 9 deaths
Exit Effect 4
• Side exits closed: 5 deaths
Exit Effect 5
• Top & bottom exits closed: 28 deaths
SEAS NRT
JSAF
SimBridge
RTI
STEALTH3D Visualization
SEAS VIS
SimBridge
Manf & Serv
Energy
Labor
Information
Finance
Infrastructure
Emergent Network Model
Consumer
EmID
S
S
S
S
S
S
E1
E4
E3
E2
E5
E6
ONA (Operational Net Assessment)
Effects
SPorts to receive messages Channels to send messages Sensors to sense messages
Modeling ExperimentationIntegration Validation VisualizationAnalysis
MRE ToolkitMRE ToolkitMRE Toolkit 6
Science BasedModels
Process/ResponseModels
Threat & RiskModels
Chemical & Bio.Models
Nuclear & Rad.Models
Cyber SecurityModels
Effects BasedModels
Nodal AnalysisModels
GISData
EconomicData
CapabilityData
LessonsLearned
ONAData
SurveillanceData
SensorData
IntelligenceData
4
High Performance ComputingDynamic Data Driven Applications
Storage, Networking, PlatformSecurity
2Cyber InfrastructureCyber InfrastructureCyber Infrastructure
OrganizationalInfrastructure
SocialEconomy
Process
Healthcare
Mixed Reality EnvironmentMixed Reality EnvironmentMixed Reality Environment
ComputationalExperimentation
Laboratory
LivingLaboratories
Mixed Reality Environment
Scaling
Validation
1
MuscatatuckE-Stadium
Borman HignwayDRCC
State of Indiana
Regenstrief Healthcare
Sensor/Instrument
Forensics
Enforcement
Intelligence
Prevention
NRP/NIMS
Technology
Processes
Training
Preparedness
HealthCare
Logistics
Organizations
Net Assessment
Response Network
5
3
Knowledge Portal-Net Assessment-Red View-Blue View-Political, Military, Social, Economic, Information, Infrastructure Summaries
Knowledge Base ONADatabase
-Effects-Nodes-Actions-Consequences-Links-Rationale
Concept
Development
Living Lab Experimentation
Computational Experimentation
Doctrine Development
Training & Education using Mixed Reality
Mixed Reality Experimentation
(Concept Scaling)
War Games (Concept
Assessment)
Concept ValidationValidated Concept
Concept Prototyping
NRP (Concept Institutionalization)
NRP (Concept Institutionalization)
Validated Scalable Concepts
Validated Scalable Concepts
Refine ConceptRefine
Concept
Refine ConceptRefine
Concept
SEAS WORLD
Bombing at the building level
SEAS CountryPopulation & Social Networks
Near Real Time Reactions and behaviors
SEAS City
SEAS City Block
SEAS BuildingSEAS BUILDING
SEAS CITYJSAF CITY
SimBridge
Agent Routing andAgent Sensing Algorithms
World Opinion
SEAS CITY BLOCK
Multi-layer, Multi-Granularity
Cities
Countries
Regions
World
Liberal View
Cultural ViewCross Cultural
View
Social Network
Realist ViewKeynsian View
X
X
X
X
X
X
X
X
X
X
X
Plurality of Thought
SEAS DDDAS Development Process
Universities
Synthetic Environment
DomainTheories
Experimental/Archived Data
Current/BestPractices Data
Lessons Learned
DoctrinalInformation
Peer-review(Validation,
Verification &Accreditation)
Expert &PractitionerValidation
VisualizationLocal CW
Province CW
National CWGlobal CW
ModelOutput
Inputs
Proprietary Data
OtherAgencies
Towards an Ecosystem
Models Bull pen
Compare theoretical behavior &
actual behavior
Accreditation for use
Open Source World
Model repository
Environment
TeamUmpire
Adversary
Behaviorally accurate, community developed, experimentally calibrated, scientifically validated, professionally verified
Tale of Two WorldsFirewall Hypothesis
Open Source World
Hidden World
Attrition World
Theories
Controls
FilterFilter
Simbridge
Synthetic Environment for Analysis and Simulation
SEAS
Crowd Formation
Coordinated Attack Simultaneous Attack
Crowd Formation
Demonstration - Counter DemonstrationDemonstration
Crowd Formation
Refugees -- Hotel Rwanda
Crowd Formation
Friday Prayer Opinion Formation
QuickTime™ and aTIFF (LZW) decompressor
are needed to see this picture.
Conclusions• Society of simulations is an effective method to link
heterogeneous simulations.– Consumers translate & interpret data specific to their needs.
• Capable of being distributed for scalable performance.– Scalable wrt. data size
pull-based, on-demand, sensing.– Scalable wrt. synchronization
asynchronous interactions, decoupling of consumer’s access patterns from producer’s.
Questions / Comments?
Shared Reality
World Region Country Province
SEAS - VIS
City Block
City
City Block Incident
City
JSAF
SEAS - NRT
GeographiesEntities