Documente Academic
Documente Profesional
Documente Cultură
Information Technology
and C4ISR
14 March 2011
• Information Assurance
• Information Management
• Complex Networks
• Signal Processing
3
FY 09/10 Research Portfolio
Finland
Cog Networking
Poland
The Netherlands Wireless Info Xfer
Information management
QIP (w/Physics PM) Czech Republic
Adversarial Planning
Network Monitoring
United Kingdom
C2 Planning
Sensor Prognostics
MAS Comm & Control Hungary
Software Dependability Data mining
Information management
Ukraine
Multimodular recurrent NNs
Spain
HPC Bulgaria
Analogical Planning
Greece
AdHoc Network Modeling
Germany
Stegonography Italy Israel
Waveform Diversity Uncertainty Management 4
Steganalysis
Information management Software regression verification
Scientific Challenges
C2 Planning & Scheduling
6
Anticipation by Analogy
7
DUAL (Kokinov, 1994)
Multi-agent Hybrid Architecture
8
Long-Term Memory and WM in DUAL
Architecture and AMBR Model
Input node
LTM WM
focus
Goal node 9
AMBR Reasoning System
glass glass
yesterday yesterday
on on
hplate hplate
kitchen kitchen
cause cause
is-broken is-broken
suicide terrorist A
suicide Kamikaze B
12
suicide Kamikaze B
Japan
motivation
Movie
“Shogun”
Honor of the
family
England
Prosperity of
the emperor
Prosperity of
Ireland
the country
13
suicide terrorist A
nostalgia
motivation
Un-satisfaction
Motivation
(CONCEPT)
14
Base “Immigrant”
Bulgarian
Ireland
nostalgia
motivation
Un-satisfaction
15
Beating his Immigrant C
wife
Stop beating
his wife
Open
Bulgarian
restaurant
Cause
Is-a Is-a
Stop bad Is-a Popularizing
actions traditional
Cause
culture
16
“Terrorist” example demonstrates
18
Scientific Challenges
Software and Systems
20
Software Regression Verification:
Proving the equivalence of similar programs
Benny Godlin Ofer Strichman
21
Functional Verification
“For every problem that you cannot solve, there is an easier problem that
you cannot solve” (George Polya, in How To Solve It) 22
Hoare‟s
1969
paper
has it all.
...
23
Project Goals
24
Notions of equivalence
1. Partial equivalence
Executions of P1 and P2 on equal inputs that terminate, and result in equal outputs
2. Mutual termination
Given equal inputs: P1 terminates P2 terminates
3. Reactive equivalence
Let P1 and P2 be reactive programs. Executions of P1 and P2 which read the same input
sequence emit the same output sequence
4. k-equivalence
Executions of P1 and P2 on equal inputs where loops and recursions are restricted to k
iterations, result in equal output
26
Proving partial equivalence
//in[A] //in[B]
A B( . . . )
B
A( . . . )
{ {
. . . . . .
//in[call A] // in[call B]
call A(. . .); call B(. . .);
//out[call A] //out[call B]
. . . . . .
} }
//out[A] //out[B]
27
Proving partial equivalence
• Let AUF , BUF be A,B, after replacing the recursive call with a call to (the
same) uninterpreted function (terminates and produce the same value on
same input).
28
29 29
Regression Verification Summary
31
Recent Transitions
• AgentFly
– ARMY funding flight tests (seeking partners)
• Adversarial Planning
– Cyberspace project fully funded by AFRL
• Barcelona Supercomputing Center
– Follow-on project funding from AFRL
• Waveform Diversity (Greco)
– Transition project in low-cost sensor array (Mike
Wickes)
32
Contact Info
James Lawton
C4ISR/IT
EOARD
Unit 4515 Box 14
APO AE 09421
33
Backup Slides
AgentFly Transitions
Deployment on PROCERUS
video
35
Agent-Based Computing in Distributed
Adversarial Planning
Michal Pechoucek, Czech Tech Univ
• Adversarial planning: a decision-making process
through which an agent constructs a sequence of
actions (possibly consisting of a single action only)
leading to the desirable goal state of the world in an
adversarial situation, i.e., when other adversarial
agents are concurrently operating in the world.
• Core Benefits:
– CPR/BDI structured plans:
• better indexing of experience
databases
• richer plan descriptions support
execution as intended Plan
– structured execution protocols STRATEGIC
• principled way of monitoring progress
and state Plan Plan Plan OPERATIONAL
• principled way of dealing deviations
Plan Plan Plan Plan
TACTICAL
38
INCA/CPR+BDI: Shared Plans
Plan
Actor Desire
Evaluation
Criterion
Activity TimePoint Belief
Resource
State-Belief
Constraint Signature Name
Capability
Precondition Parameter
Method
Effect
Annotation
Activity has-a
is-a
Backup Slides
Multcore Programming
“Programming models for heterogeneous
multicore systems”
Jesus Labarta, Barcelona Supercomputing Center
• Background: To efficiently use the multicore machines we need:
– Understanding of application behavior
– Programming models that can express apps in a general/abstract/portable way.
– Intelligent compilers and runtime environments capable of executing very efficiently those
applications on any target machine.
• Project Objectives:
– Refine the StarSs programming model developed at BSC for multicore processors.
– Apply performance analysis tools developed at BSC to better understand parallelism in AFRL-
provided example applications
• Project Approach:
• Programming model:
– Extend dependency mechanism
– Add support for new core/device type (GPUs)
– Handling of partially aliased and strided references
– Hybrid MPI/SPMSs
• Performance analysis:
– Analyze and improve performance of AFRL-provided
apps
– Determined that tools can be extended with pattern
matching techniques 41