Bonneau - Complex Networks Foundations of Information Systems - Spring Review 2013

DISTRIBUTION A: Approved for public release; distribution is unlimited.
Integrity  Service  Excellence
Complex Networks/
Foundations of
Information Systems
6 March 2013
Robert J. Bonneau, Ph.D.
Division Chief
AFOSR/RTC

NAME: Complex Networks/Foundations of Information Systems
BRIEF DESCRIPTION OF PORTFOLIO:
Complex Networks and Foundations of Information Systems uses measured information
to assure, manage, predict, and design distributed networks, systems, and architectures
2013 Spring Review
• Local Network Research: Guarantee and assure information transmission
• Network Management Research: Network and system protocols for resilient and
robust
• Global Network Research: Mathematically represent network performance and
design robustness
• Foundations of Information Systems Research: Measure, predict, and verify
information system properties
LIST SUB-AREAS IN PORTFOLIO:

Goals:
• Measure system performance and calculate risk
• Preserve critical information structure and minimize latency over a
heterogeneous distributed network and system
• Ensure network and system robustness and resilience under a diverse set of
resource constraints and manage using dynamic models
• Represent global properties of a networked system in a unified
mathematical framework for architecture and design
• Assess and Predict heterogeneous distributed systems performance using
unified mathematical framework
Complex Networks &
Foundations of Information Systems
Foundations Goals Complex Networks Goals

Complex Networks &
Foundations of Information Systems
Payoffs:
• Preserve critical information structure in a network rather than just delivering
packets or bits
• Quantify likelihood of a given network management policy to support critical
mission functions
• Predict and manage network and system failure comprehensively
• Assess and verify properties of a distributed heterogeneous system where
there is limited access to its elements
• Assess dynamic Air Force system mission performance and assess risk of
failure

Complex Networks and Information Systems uses measured information to assure,
manage, predict, and design distributed networks, systems, and architectures
Complex Networks and
Information Systems Roadmap
Heterogeneous Dynamic
Air Force Systems
Global System
Local
System
System
Management Critical Information
System Analysis
& Prediction
Information Systems
Research
(System Measurement
& Assessment)
Diverse
Types of Systems
Local Network/Systems
Research
Global Network/Systems
Research
Complex Networks
(Unified Representation, Design,
Architecture)
Preserve
Critical
Information
Network/Systems
Management Research
System
Measurement

What to measure?
How to measure?
Information Systems
Measurement
We wish to understand what and how to measure systems for representations that
are traceable to mission performance and have quantifiable risk

System Measurement
and Probabilistic Model Checking
Ed Clarke, CMU, David Musliner, SIFT
Approach: Techniques on what and how to measure on any software, network, and
hardware integrated system with probabilistic model checking.
Payoff: Large system state spaces can be verified and risk calculated in a computationally
efficient way.
Dynamic System
Model of Critical System
Parameters
Heterogeneous
Systems
Toward Online
System Analysis With Measurement
Probabilistic System Prediction and
System Risk Analysis
System Probability Logic
State Using Geometric
Invariants

Dan Speilman, Yale
Accurate Measurement of
Systems Graph Parameters
Approach: There are many different ways to represent networked systems. There needs to be
a principled way of selecting the best inputs for geometric models to parameterize system
performance.
Payoff: Systems can be parameterized in such a way as to minimize computational overhead
and inaccuracies in model prediction using sparse geometric graph theory.
MacArthur Genius Award Winner 2012
Critical Systems
Information
Sparse Regression of Graph
To Build Best Geometric
Model
Best Model?
Best System
Parameterization

Statistical System
Estimation and Risk Analysis
Eric Kolaczyk, Boston University
Approach: Sample large systems guided by understanding of geometric invariants and risk
analysis in how measurements were constructed.
Payoff: Minimum risk can be incurred from measurements of large system while still
sampling in most computationally efficient way.
Minimum Risk Outcome for Graph Sampling Geometric Strategies for Sampling Graph
Risk Performance Peer to Peer Network Data

Less: Information Loss Under
Disruption
More: Latency, Resource
Intensive
Less: Latency/Disruption
Tolerant
More: Controllable
Information
Sources
Information
Timescale ∆t
Architecture Information
Distribution
Architecture
Information Loss
With Interference
Architecture
Information
Recovery
• We wish to characterize network and system performance from measurement and
develop coding, protocol, and architectures that adapt according to network/system state
Complex Networks
System Representation
Deterministic
Routing
(ex: Core/Backbone)
Random Network
(ex: Mobile Ad Hoc)
Hybrid Network
(Mesh)
Deterministic
Routing
(ex: OSPF)
Hybrid Routing
(ex: OLSR)
Random Protocol
(ex: Flooding)
Hybrid Code
(ex: Network Code)
Random Code
(ex: Rateless Code)Packets
Packets Groups
Packets Blocks
Deterministic Code
(eg. Trellis Code)
Source 1
Source 2
Source 3
Source 1
Source 2
Source 3
Information
Receivers
Algorithm State Adapts
According to
Conditions (Feedback)

Network
Taxonomy
Approach: Develop a taxonomy of network properties through curvature and timescale
invariants from hyperbolic geometry
Payoff: Space of network models can be developed and design principles and analysis
algorithms derived to predict network behavior
Narayan, Saniee UC Santa Cruz/Lucent
Social Network
Application
Protocol
Architecture
Taxonomy of Physical Architecture
Across Networks
Curvature vs. Timescale and
Network LayerGeometric Curvature Invariant
Characterization of
Network
Stack Across
Layers
(curvature connected
to congestions)

Entropy Classes for Dynamic
Network Models
Approach: Deriving models for network protocol analysis has assumed fixed protocols
without the benefit of feedback or dynamic correlations in coding and protocol. Using Renyi
correlation analysis and entropy to model this wider class of models allows greater flexibility
in describing statistical classes.
Payoff: Complex phenomena such as feedback in coding can be modeled under dynamic
heterogeneous conditions.
Lizhong Zheng, MIT
Renyi Channel Correlation Analysis
(connected to geometric curvature)
Network Channel Architecture
Feedback Channel & Geometric Invariants

Approach: Coding theory classically assumes fixed statistic and fixed methods of
coding for a given set of network and interference conditions
Payoff: Feedback in coding methods can significantly improve throughput under
dynamic network and interference conditions
D. Tse, Berkeley, P. Gupta, Alcatel Lucent, D. Shah, MIT
Feedback in Coding Theory
Feedback Coding Approach Areas Where Improves Performance
(including side information)

Information Systems
Assessment
Hybrid Architecture
Hybrid Protocol
Hybrid Content
Random Architecture
Random Protocol
Random
Content
Deterministic
Architecture
Deterministic
Protocol
Deterministic Content
Measured Performance RegionsHeterogeneous
Information
Network States
(packets, packet blocks,
packet groups)
Software States
(variable, subroutine,
program)
Hardware States
(register, ram, virt. mem)
System
Measurements
Less: Information Loss Under
Disruption/Live
More: Latency, Resource
Intensive/Safe
Less: Latency/Disruption
Tolerant/Safe
More: Controllable/Live
Best Integrated
Performance Region
(timescale/level
of abstraction)
Global
Properties
Statistical
Properties
Stable/Resourced
SecureUnstable/Un-resourced
Insecure
Predict system performance using mathematical invariants from measured data
Invariants Predict
Performance

ST Yau, Harvard, Ali Jadbabaie, UPenn, Fan Chung Graham, UCSD
Hodge Theory for Invariants in Network Algorithms
Approach: Hodge decompositions allow natural functional invariants to be used on
extremely complex structures
Payoff: Analysis of Hodge decomposition can be applied to whole classes of
networks to characterize invariant parameters and predict performance.
Hodge Decomposition
Gives Powerful Functional Invariants
Invariant Analysis of
Page Rank (Google)
Invariant Analysis of
Google’s Page Rank Algorithm
for Different Values of Beta

John Harer/Sayan Mukherjee Duke Konstantin Mischaikow/Rutgers, Samson Abramsky Oxford, Peter Chin Johns Hopkins
Approach: Dynamic measured data has correlated structures that are difficult to
characterize by standard tools in algebraic topology such as homology. Additional features
such as Conley index can augment invariants provided by algebraic topology.
Payoff: Dynamic modes of networks and networked systems can be captured by new
methods in discrete Morse theory and algebraic topology to predict performance
Topological Invariants for Model Checking
Network & System Dynamics
Dynamic Time Series
Initial Time Series
Graphs from
Time Series
Graphs With
Invariant Cycle
Time Series Data
Use of Conley Index Allows
Verification of Dynamics for System Assurance
?
Geometric
Invariant
(homology)
Model Equivalence Checking
A
B
Bugs and System Instabilities Can Be Measured and Model Checked in Real Time

Categorical Invariants on
Information Flows
Rob Ghrist, UPenn
Approach: Dynamic network information flow can be parameterized by its max flow
min cut value relative to the network topology
Payoff: Categories of flows can be developed using sheaf theoretic formulation for
max-flow min-cut parameters to characterize and verify system performance
Sheaf Formulation For Max Flow Min Cut
and Network Code Flow Parameters Cut Parameters Max Flow =
Min Cut Invariant
Categorical Invariants For Coded Information Flows
Corresponding Network Flow Conditions
Joint project with DARPA/DSO GRAPHS program

Complex Networks Transition Activities
Complex Networks uses advanced mathematical analysis of information systems
measurements to resource, verify, and secure distributed Air Force infrastructures
- Princeton MURI – dynamic analysis of packets for airborne network resource management
- Yale/Telcordia commercial grant – real time system security policy verification
- Columbia/MIT – information theory for new quantum semiconductors
New Systems Components in
Architecture
Cloud Security/Cyber
Airborne Networks
Resource Analysis Scheduling
Geometric Analysis
Of Distributed System
Resources Dynamics
Intelligent Automation
STTR
AFRL
Lab Tasks
University
Research
Transition
Mechanisms
DARPA InPho
Information Theory for
Quantum
Semiconductors
Cloud Analysis
Using Operator
Algebras

Space and Airborne
Cognitive RF
Approach: Use advanced non-parametric geometric measurement and game theoretic
techniques with Generalized Likelihood ratio test to enable training on RF interference data in
conjunction with advanced MIMO beam-forming techniques to remove interference
Payoff: AF will be able to conduct operations in crowded dynamic spectrum environment
Scott Erwin, AFRL/RV, John Matyjas, AFRL/RI, Vasu Chakravarthy, AFRL/RY
Space Time Network Adaptive Detection
Network
Measurement
& Invariants
Complex RF Environment
5 10 15 20 25 30 35
10
- 6
10
- 5
10
- 4
10
- 3
10
- 2
10
- 1
10
0
Transmit power to noise ratio, P/N0
, dB
BitErrorRate(BER)
BPSK signals, CDMA (L = 16), ds, d
= 100m, UAV height = 1Km
Direct transmissions
Two-hop relaying
Cooperative transmissions
Direct MIMO: [2x1] Alamouti scheme
Improved ThroughputImproved Detection

Quantum Network & System
Measurements
Jeff Yepez/AFRL/RD, Paul Alsing/AFRL/RI, Atilla Szep, AFRL/RY
Data from Quantum
Network & System
Measurements
Photonic Device
Measurements Optical
Measurements
Approach: Verification of classical vs. quantum states in a network provides a new class of
measurement problems which need invariants for analysis of states of large scale quantum
systems for information assurance, communication, and computing
Payoff: Integrating new invariant network techniques with new quantum representations will
allow a much more integrated way of analyzing large scale quantum network and
computational systems
Representation of Quantum Logic
New MURI 2012: Measurement and Verification in Quantum Information Systems

2D Ricci flow
Approach: Verification of classical and quantum states in a network and system provides a new class of
measurement problems which need invariants such as computational homology for analysis of states
Payoff: Integrating new invariant network techniques with new classical and quantum representations
will allow a much more integrated way of analyzing & securing large scale networked and computational
systems
Computational Homology and Ricci
Flow for Verification of Classical and Quantum States
Paul Alsing/AFRL RI, Howard Bramson/Syracuse, Warner Miller, FAU
Network/System Decomposition
Network Information
Properties Of Curvature

Approach: Complex Networks invariants can be used to assess risk on human vs. machine operation
when faced with complex decision task.
Payoff: Detailed understanding and automation of how to trade human vs. machine performance in time
critical functions such as cyber security, autonomous operation of air vehicles
Warfighter
Machine Learning
On Computational Infrastructure
Performance and
Representation
Estimate Decision Risk
Using Complex Network Invariants
Measurement
Machine Learning Risk
Lower
Human Risk
Lower
Complex Networks Invariants for Risk Assessment
Measurement of Human vs. Machine Performance
Leslie Blaha, AFRL/RH
Network RepresentationHidden Variable
Theory
Human
Behavior

Recent Program Awards
• Dan Speilman
– MacArthur Genius Award (2012)
• Ingrid Daubechies
– Fellow of the American Mathematical Society 2012
– Benjamin Franklin Institute Medal, 2012
– Leroy Steel Prize From American Mathematical Society 2012
– IEEE Jack Kirby Signal Processing Medal 2011
• Robert Calderbank
- IEEE/Hamming Medal Winner (2012)
• Vincent Poor
- National Academy of Science (2011)
• Emmanuel Candes:
– Collatz Prize (Mathematics) , (ICIAM) (2011)
– Winner, Sixth Vasil A. Popov Prize (Mathematics) ,(2010)
• ST Yau
– Wolf Prize Mathematics, 2010
• Joel Tropp:
– Eighth Monroe H. Martin Prize, 2011
• Yonina Eldar
– IEEE Fellow 2013
• Mung Chiang:
– IEEE Kiyo Tomiyasu Award in 2012
• Junshan Zhang

Academia/Commercial Outreach
• Keynote Lecture, American Society of Mechanical Engineers, Complex
Systems 2012
• Keynote Lecture, International Conference on Complex Networks,
2012
• Keynote Lecture, IEEE, CogSima, 2013
• Invited Lecture, Yale Mathematics, Anniversary of Coifman, Jones,
Rokhlin Achievements
• Organizer: London Institute of Mathematics: Mathematics of on
Statistical Verification, March 2012
• Organizer, OSTP Meeting on Complex Engineered Systems, 2012
• Invited Speaker, Allerton, IEEE Conference on Information Theory,
University of Illinois 2012
• Invited Speaker, IDGA Big Data, Federal Meeting Washington DC, 2013
• Invited Speaker, Neuro-Information Processing, Lake Tahoe, 2012
• Invited Speaker: Dagstuhl Germany, Mathematics of Information Flow,
2012
• Invited Speaker: USCD Information Theory and Applications, LaJolla
2013

Program Impact &
Collaboration with Agencies
• OSTP/NITRD – Co-Chair Large Scale Networks Working Group
- New national thrust – Complex Networks and Systems inspired by AFOSR
program – Workshop “Complex Engineered Networks” organized by leader of
AFOSR Complex Networks MURI “Information Dynamics in Networks”
• ASDR&E
- Engineered Resilient Systems – Complex Networks and Foundations of
Information Systems on Roadmap
• DARPA Collaboration/joint program reviews
– Graphs – Mathematics of graphs and networks agent
– Defense Science Office Mathematics Advisory Panel
– InPho – Information in a photon/quantum network collaborative funding
– Com-EX cognitive network program transition agent
• IARPA – Quantum Computer Science Working Group
• ARL/ARO Network Science Board of Advisors
• NSF Future Internet, Cyber Physical Systems

Complex Networks Trends
• Local Network Theory
– Geometric and binary information coding
– Coding information with network performance objectives
– Integration with verification and quantum methods
• Network Management
– Nonparametric strategies for assessing network performance
– Distributed strategies for measuring and assessing network information
transfer
– Sparse network management
• Global Network Theory
– Geometric flow analysis for prediction and management of network
performance
– Global state space taxonomy and categorization
• Information Systems Research
- Combined network, software, and hardware analysis
- Defining correct input data for given mathematical assessment
- Invariant metrics for analysis of network performance

Cyber Operations: Joint University Center of Excellence:
“Cyber Vision 2025” – Enabling Technologies workshops
“Secure Cloud Computing” with university and AFRL/RI
Dynamics and Control : Verification and Validation of Complex Systems
Physics and Materials: New Joint MURI Topic: “Large Scale Integrated
Hybrid Nanophotonics”
Socio-Cultural Analysis: Social Networks – Joint MURI Topic: “Stable
Metrics for Inference in Social Networks ” – UCLA/USC/ASU
Quantum: Interaction with quantum network and quantum estimation
processes through lab tasks
- Joint EOARD initiative at Cambridge
Information Fusion: Critical feature selection in sensor networks
Optimization: Competing optimization requirements.
Decision: Networks of neurons.
Biology: Systems biological processes as networks.
Other Program Interactions

Transition Activities
• AFRL
– AFRL/RI – Cyber Vision 2025 workshops/Illinois Center for Secure
Cloud Computing/
– AFRL/RI/RY – DARPA InPho program/DARPA Graphs program
– AFRL/RI/RY/RV – distributed secure space communications
– AFRL/RW/RV/RH – verification and validation of complex systems
• STTR
– Intelligent Automation: Transition to ESC of Airborne Networks
management – transition to Boeing for test-bed
– Avirtek: Secure router application interface- AFRL/RI
– Andro – Joint Spectrum Center Lockheed transition of automated
spectrum management tool

Transition Activities
• Customer/Industry
– Collaboration with ACC/GCIC, Air Force Spectrum Management
Agency on JALIN ICD
– Collaboration with Boeing, ESC, IAI for transition of coding and
routing management protocols baseline CORE tools to Rome
Lab for possible integration in CABLE JCTD
– Briefing to Space Command/Peterson for potential collaboration
– Interaction with Northrop Grumman/BACN airborne networking
program for potential collaboration
• OSD
– Complex Systems Engineering and Systems 20/20 initiative
– Software Assurance and Security Initiative
– Robust Command and Control Initiative
• Commercial
– Interaction with Stanford on real time network information
recovery
– New initiatives with Akamai for content distribution analysis
– Interaction with USFA/DHS/CISCO on router algorithm design

Bonneau - Complex Networks Foundations of Information Systems - Spring Review 2013

More Related Content

What's hot (19)

Viewers also liked (20)

Similar to Bonneau - Complex Networks Foundations of Information Systems - Spring Review 2013 (20)

More from The Air Force Office of Scientific Research (20)

Recently uploaded (20)

Bonneau - Complex Networks Foundations of Information Systems - Spring Review 2013