flatland visualization of a decision support tool …...portfolio for the set of s&t...
TRANSCRIPT
Flatland Visualization of A Decision Support Tool
Architecture
Thomas Preston CaudellDepartment of ECE
University of New [email protected]
The Team
• New Mexico State University- Jim Cowie, Chris Fields, Hung Nguyen , Bill Ogden, Ram Prasad
• Monterey Institute for International Studies-Gary Ackerman, Markus Binder, Sundara Vadlamudi
• University of New Mexico- Frank Gilfeather, Thomas Caudell, Panaiotis, Tim Ross, Mahmoud Taha
The Problem
Allocation of Science & Technology (S&T) research funding to maximally reduce the threat and consequences of CB attacks on critical assets is complex and very hard to optimize globally.
Design Goals
• Develop an analytic and algorithmic framework for threat consequence minimization.
• Create a feasible system architecture to evaluate modeling, analysis approaches, and user interactions within this framework.
• Enhance decision process transparency.
Initial ArchitectureNo Temporal Dynamics
$ $ $ $ $ $ $ $
$
Calculated consequence vector for each threat scenario (simor experts).
Individual threat scenarios derived from tree.
User adjustable likelihood for each threat scenario.
Total S&T investment funds available for this analysis.
User adjustable funding portfolio for the set of S&T remediations.
Calculated net consequence for this set of scenarios, likelihoods, remediationsand funding portfolio choices.
User selected possibility tree for one class of threat scenarios.
Calculated effectivityof a specific remediation against a specific threat scenario, a function of investment
R R R R R R R R
Set of S & T remediaitons
C
Specification of Scenarios
• Possibility Trees • Spanning sets of scenarios• Vectors of consequences per scenario• Possible continuous scenario space• Possible continuous consequence space
Creating Models of Consequences, Costs, and Effectiveness
• Relate remediation funding level to its effectiveness against a given scenario’s consequences.
• Scientific Simulations• Machine learning models• Knowledge based systems
Details of ArchitectureConsequence Flow Model
Scenario(k) conseq(k) likelihood(k) effectivity(k,m)
$(m)
remediation(m)
Σ ExpectedConsequence
effectivity(k,n)
Scenario(i) conseq(i) likelihood(i) effectivity(i,m) effectivity(i,n)
$(n)
remediation(n)
$ Total
Mockup Mathematical ModelAssumes only one type of consequence per scenario.
NumScenarios NumRemediations
Expected Conseq = α Σ conseq(k) * likelihood(k) * Π (1-effectivity(k,m)) k=1 m=1
Where:∗ α is a normalization constant, * conseq(k) is a real scalar, * likelihood(k) is a probability (sums to 1), * $(m) is a real (sums to $ Total = 1),* effectivity(k,m) = β(k,m) * F($(m)), ∗ β(k,m) = rand(0,1).
User Adjustables:a) {likelihood(k), k=1, NumScenarios} b) {$(m), m=1, NumRemediations}
Optimization Loop
$ $ $ $ $ $ $ $
$
R R R R R R R R
C
{ }{ }
Optimization / Ranking Tools
R R$
Tree
LikelihoodsRemediationsTotal budget
(C, , … , …. , )$ $
(C, , … , …. , )$ $
(C, , … , …. , )$ $
(C, , … , …. , )$ $
Model
Model
Input Parameters Analysis Framework Ranked Consequences
OptimizationAllocation of funds to minimize expected consequences
Optimization / Ranking Tools (C, , … , …. , )$ $
(C, , … , …. , )$ $
(C, , … , …. , )$ $
(C, , … , …. , )$ $
{ }{ }R R
$
Likelihoods
Remediations
Total budget
ModelTree
Input Parameters Analysis Framework Ranked Consequences
Mockup: $n(k+1) = $n(k) - eta * gradient( Expected Consequences)
Temporal Dynamics
Scenario(k) conseq(k) likelihood(k) effectivity(k,m)
$(m)
remediation(m)
Σ ExpectedConsequence
effectivity(k,n)
Scenario(i) conseq(i) likelihood(i) effectivity(i,m) effectivity(i,n)
$(n)
remediation(n)
$ Total
The visualization team
• Shan Xia - UNM, ECE
• Victor Vegara- UNM, ECE
• Panaiotis- UNM, ECE & Music
• Steve Smith- LANL
• Thomas Caudell- UNM, ECE & CS
Features of visualization• Goal: transparency into computational model and
decision process.• Consequence-flow metaphor• Real-time user adjustable parameters• User viewpoint control to manage complexity• Drill-down for more details• Animation of calculations and optimization• Complementary sound representation of system
states and dynamics• Implemented in Flatland.
Flatland: modular applications
Whiteboard
Collaboration
Foundational Modules
Base EnvironsHCIModules
Flatland Core and Services
AG Interfaces
Medical Models
AI
Haptics
Database Interfaces
NLU HPC Interfaces
KB KB KB
ML ML
Visualization of Mockup System
Scenario TreeConsequences per Scenario
Likelihood of Scenario
Effectivity Matrix
Funding PortfolioRemediations
Expected Consequences
Future advancements
• Integration with models and optimization,
• Integration with scenarios,• Scalable scenario & remediation
representations,• Effectivity model representations,• Quantitative measures of
performance.
Invitation to demonstration at the
University of New MexicoVisualization Laboratory
Center for High Performance Computing
- Thursday -
Contact: [email protected]