20
ELECTRICAL ENGINEERING DEPARTMENT 1 SPTM Expert Panel SPTM Expert Panel SPTM Expert Panel SPTM Expert Panel In Network Processing, Adaptation, and Learning In Network Processing, Adaptation, and Learning In Network Processing, Adaptation, and Learning In Network Processing, Adaptation, and Learning ICASSP 2011 ICASSP 2011 ICASSP 2011 ICASSP 2011 May 27, 2011 Prague, Czech Republic http://www.istockphoto.com/ A. H. Sayed

SPTM Expert Panel - IEEE Signal Processing Societysignalprocessingsociety.org/uploads/Publications/SPM/... · 2018-12-21 · SPTM Expert Panel In Network Processing, Adaptation, and

  • Upload
    others

  • View
    6

  • Download
    0

Embed Size (px)

Citation preview

Page 1: SPTM Expert Panel - IEEE Signal Processing Societysignalprocessingsociety.org/uploads/Publications/SPM/... · 2018-12-21 · SPTM Expert Panel In Network Processing, Adaptation, and

ELECTRICAL ENGINEERING DEPARTMENT 1

SPTM Expert PanelSPTM Expert PanelSPTM Expert PanelSPTM Expert Panel

In Network Processing, Adaptation, and LearningIn Network Processing, Adaptation, and LearningIn Network Processing, Adaptation, and LearningIn Network Processing, Adaptation, and Learning

ICASSP 2011ICASSP 2011ICASSP 2011ICASSP 2011May 27, 2011

Prague, Czech Republic

http://www.istockphoto.com/

A. H. Sayed

Page 2: SPTM Expert Panel - IEEE Signal Processing Societysignalprocessingsociety.org/uploads/Publications/SPM/... · 2018-12-21 · SPTM Expert Panel In Network Processing, Adaptation, and

ELECTRICAL ENGINEERING DEPARTMENT 2

BIRD SWARMSBIRD SWARMSBIRD SWARMSBIRD SWARMS

(Denmark)(Denmark)(Denmark)(Denmark)

BBC videoBBC videoBBC videoBBC video (requires Internet connection)

Page 3: SPTM Expert Panel - IEEE Signal Processing Societysignalprocessingsociety.org/uploads/Publications/SPM/... · 2018-12-21 · SPTM Expert Panel In Network Processing, Adaptation, and

ELECTRICAL ENGINEERING DEPARTMENT 3

FISH SCHOOLSFISH SCHOOLSFISH SCHOOLSFISH SCHOOLS

BBC Video #2BBC Video #2BBC Video #2BBC Video #2

BBC Video #3BBC Video #3BBC Video #3BBC Video #3

BBC Video #1BBC Video #1BBC Video #1BBC Video #1

(videos require

Internet connection)

Page 4: SPTM Expert Panel - IEEE Signal Processing Societysignalprocessingsociety.org/uploads/Publications/SPM/... · 2018-12-21 · SPTM Expert Panel In Network Processing, Adaptation, and

ELECTRICAL ENGINEERING DEPARTMENT 4

V-formationhttp://www.flickr.com/photos/gillesgonthier/2510118726/

BIRD FORMATIONSBIRD FORMATIONSBIRD FORMATIONSBIRD FORMATIONS

U-formationwww.amazingtopics.com

BBC videoBBC videoBBC videoBBC video (requires Internet connection)

Page 5: SPTM Expert Panel - IEEE Signal Processing Societysignalprocessingsociety.org/uploads/Publications/SPM/... · 2018-12-21 · SPTM Expert Panel In Network Processing, Adaptation, and

ELECTRICAL ENGINEERING DEPARTMENT 5

BACTERIA MOTILITYBACTERIA MOTILITYBACTERIA MOTILITYBACTERIA MOTILITY

Press article:

Scientists harness

bacteria to turn

microscopic gears(Argonne National Lab, Northwestern, Princeton)

Research published in Proceedings

of the National Academy of Scienceshttp://www.clker.com/

play videoplay videoplay videoplay video

Page 6: SPTM Expert Panel - IEEE Signal Processing Societysignalprocessingsociety.org/uploads/Publications/SPM/... · 2018-12-21 · SPTM Expert Panel In Network Processing, Adaptation, and

ELECTRICAL ENGINEERING DEPARTMENT 6

RATIONAL PATTERNS OF BEHAVIORRATIONAL PATTERNS OF BEHAVIORRATIONAL PATTERNS OF BEHAVIORRATIONAL PATTERNS OF BEHAVIOR

While each individual agent in an animal group is

not capable of complex behavior, it is the

combined coordinationcombined coordinationcombined coordinationcombined coordination among multiple agents

that leads to the manifestation of complex

patterns of behavior.

Interactions among Interactions among Interactions among Interactions among

unsophisticated agentsunsophisticated agentsunsophisticated agentsunsophisticated agents

Sophisticated, organized, and Sophisticated, organized, and Sophisticated, organized, and Sophisticated, organized, and

rational patterns of behaviorrational patterns of behaviorrational patterns of behaviorrational patterns of behavior

LEAD TO LEAD TO LEAD TO LEAD TO

HOW AND WHY?HOW AND WHY?HOW AND WHY?HOW AND WHY?

http://www.visualcomplexity.com

Page 7: SPTM Expert Panel - IEEE Signal Processing Societysignalprocessingsociety.org/uploads/Publications/SPM/... · 2018-12-21 · SPTM Expert Panel In Network Processing, Adaptation, and

ELECTRICAL ENGINEERING DEPARTMENT 7

IMPORTANT QUESTIONSIMPORTANT QUESTIONSIMPORTANT QUESTIONSIMPORTANT QUESTIONS

How and why organized behavior arises at the

group level from interactions among

agents withoutwithoutwithoutwithout central control?

What communication topologies enable such

behavior? How can different topologies be

justified? What type of processing and algorithms

exhibit this type of intelligence?

How is information quantized? How does

information propagate through the network?

How does mobility contribute to learning?

Page 8: SPTM Expert Panel - IEEE Signal Processing Societysignalprocessingsociety.org/uploads/Publications/SPM/... · 2018-12-21 · SPTM Expert Panel In Network Processing, Adaptation, and

ELECTRICAL ENGINEERING DEPARTMENT 8

SIGNAL PROCESSING TOOLSSIGNAL PROCESSING TOOLSSIGNAL PROCESSING TOOLSSIGNAL PROCESSING TOOLS

Distributed processing, filtering, and estimation.

Adaptation, learning, and tracking.

Real-time processing.

DISCIPLINESDISCIPLINESDISCIPLINESDISCIPLINES: Ecology, Evolutionary Biology, Animal Behavior, Physics, Biology, Computer Science.

Page 9: SPTM Expert Panel - IEEE Signal Processing Societysignalprocessingsociety.org/uploads/Publications/SPM/... · 2018-12-21 · SPTM Expert Panel In Network Processing, Adaptation, and

ELECTRICAL ENGINEERING DEPARTMENT 9

DIFFUSION STRATEGIES DIFFUSION STRATEGIES DIFFUSION STRATEGIES DIFFUSION STRATEGIES (example)

Combine-then-Adapt Diffusion LMS

(node k, time i)

Processing (one example)

Merging/Combination11

kk

22

N nodes

Neighborhood

Page 10: SPTM Expert Panel - IEEE Signal Processing Societysignalprocessingsociety.org/uploads/Publications/SPM/... · 2018-12-21 · SPTM Expert Panel In Network Processing, Adaptation, and

ELECTRICAL ENGINEERING DEPARTMENT 10

AGENTAGENTAGENTAGENT----BASED MODELS (ECONOMICS)BASED MODELS (ECONOMICS)BASED MODELS (ECONOMICS)BASED MODELS (ECONOMICS)

``An agent-based model is a computerized simulation of a number of decision-makers (agents) and institutions, which interact through prescribed rules… Suchmodels do not rely on the assumption that the economy will move towardsa predetermined equilibrium state, as other models do. Instead, at any given time, each agent acts according to its current situation, the state of the world around it and the rules governing its behaviour. .. Agent-based simulations can handle a far wider range of nonlinear behaviour than conventional equilibrium models.’’

Page 11: SPTM Expert Panel - IEEE Signal Processing Societysignalprocessingsociety.org/uploads/Publications/SPM/... · 2018-12-21 · SPTM Expert Panel In Network Processing, Adaptation, and

ELECTRICAL ENGINEERING DEPARTMENT 11

Cognitive networks consist of spatially distributed

nodes linked by a connection topology.

ADAPTIVE NETWORKSADAPTIVE NETWORKSADAPTIVE NETWORKSADAPTIVE NETWORKS

The nodes are generally isotropic without any

particular node taking a central control role.

Page 12: SPTM Expert Panel - IEEE Signal Processing Societysignalprocessingsociety.org/uploads/Publications/SPM/... · 2018-12-21 · SPTM Expert Panel In Network Processing, Adaptation, and

ELECTRICAL ENGINEERING DEPARTMENT 12

The nodes cooperate with each other through

local interactions.

ADAPTIVE NETWORKS: ADAPTIVE NETWORKS: ADAPTIVE NETWORKS: ADAPTIVE NETWORKS: Local InteractionsLocal InteractionsLocal InteractionsLocal Interactions

The nodes adapt their behavior in response to

data collected at the nodes and from neighbors.

Page 13: SPTM Expert Panel - IEEE Signal Processing Societysignalprocessingsociety.org/uploads/Publications/SPM/... · 2018-12-21 · SPTM Expert Panel In Network Processing, Adaptation, and

ELECTRICAL ENGINEERING DEPARTMENT 13

The diffusion and sharing of information results

in a form of collective intelligence relative to

non-cooperative networks.

ADAPTIVE NETWORKS: ADAPTIVE NETWORKS: ADAPTIVE NETWORKS: ADAPTIVE NETWORKS: Complex BehaviorComplex BehaviorComplex BehaviorComplex Behavior

Page 14: SPTM Expert Panel - IEEE Signal Processing Societysignalprocessingsociety.org/uploads/Publications/SPM/... · 2018-12-21 · SPTM Expert Panel In Network Processing, Adaptation, and

ELECTRICAL ENGINEERING DEPARTMENT 14

FISH SCHOOL AVOIDING OBSTACLESFISH SCHOOL AVOIDING OBSTACLESFISH SCHOOL AVOIDING OBSTACLESFISH SCHOOL AVOIDING OBSTACLES

Page 15: SPTM Expert Panel - IEEE Signal Processing Societysignalprocessingsociety.org/uploads/Publications/SPM/... · 2018-12-21 · SPTM Expert Panel In Network Processing, Adaptation, and

ELECTRICAL ENGINEERING DEPARTMENT

PREDATORS + PREYPREDATORS + PREYPREDATORS + PREYPREDATORS + PREY

15

Page 16: SPTM Expert Panel - IEEE Signal Processing Societysignalprocessingsociety.org/uploads/Publications/SPM/... · 2018-12-21 · SPTM Expert Panel In Network Processing, Adaptation, and

ELECTRICAL ENGINEERING DEPARTMENT 16

CHALLENGESCHALLENGESCHALLENGESCHALLENGES

Modeling of complex behavior over dynamic

networks: adaptation, cognition, mobilityadaptation, cognition, mobilityadaptation, cognition, mobilityadaptation, cognition, mobility.

What kind of signal processing and signal processing and signal processing and signal processing and

information sharinginformation sharinginformation sharinginformation sharing enables organization and

rationality over cognitive networks?

Design systems that emulate self-

organization structures and are able to learn

and adapt on the fly � can solve complex can solve complex can solve complex can solve complex

problemsproblemsproblemsproblems....

Page 17: SPTM Expert Panel - IEEE Signal Processing Societysignalprocessingsociety.org/uploads/Publications/SPM/... · 2018-12-21 · SPTM Expert Panel In Network Processing, Adaptation, and

Trends in Signal Processing Theory and Methods

Vikram Krishnamurthy

Department of Electrical and Computer EngineeringUniversity of British Columbia

[email protected]

May 26, 2011

Page 18: SPTM Expert Panel - IEEE Signal Processing Societysignalprocessingsociety.org/uploads/Publications/SPM/... · 2018-12-21 · SPTM Expert Panel In Network Processing, Adaptation, and

An incomplete list of interesting areas.....

SSP on dynamic graphs: Consensus, games, information flows,social networks. [Society, sensor networks]

High dimensional SP and sparsity: Concentration of measuretheorems e.g. Compressed Sensing – lots of papers in ICASSP;TSP, 2 special issues of JSTSP. [Machine Learning, Biomedical]

Financial time series and SP [Finance]

Smart grid – SP magazine, ICASSP special session [Energy]

Distributed Sensing, Decision Making and Control Cognitivesystems, social learning, Bioinspired SP, dynamical games.[Economics, defence, energy efficiency]

SP and stochastic simualation in life sciences (from Molecularbiology to biomedical imaging and instrumentation).nature.com website: Pfizer, Novartis. SimBiologyTM extendsMATLAB¨ with tools for modeling, simulating, and analyzing biochemicalpathways. SimBiology lets you simulate a model using stochastic ordeterministic solvers and analyze your pathway with tools such asparameter estimation and sensitivity analysis.

Page 19: SPTM Expert Panel - IEEE Signal Processing Societysignalprocessingsociety.org/uploads/Publications/SPM/... · 2018-12-21 · SPTM Expert Panel In Network Processing, Adaptation, and

Sensor-Adaptive Signal Processing – Economics viewpoint

Noise

Signal

SensorSignal Processing

Estimate

Feedback (Stochastic Control)

Topic 3: Global Game for

Sensor Activation

! Gives a communication free

deployment protocol.

! Simple threshold policies are in Nash

equilibrium.

! If uncertainty is high or congestion is low.

! Compare with other approaches:

! Local communication & self-organization

[Biswas & Phoba 06, Clare & Pottie 99,

Aroraa et. al. 04]

! Flocking

!"#$%&'$()*"+(),-.

/*)&

!-#,01#&

How can agents autonomously manage their behavior?Game theory can be used as a synthesis and analysis tool. How cansimple behavior by agents result in sophisticated global behavior? How toachieve consensus in decisions? (S. Hart, Econmetrica, 2005).

How can agents learn from actions of other agents?Social learning results in herding behavior.In 1995, Treacy & Wiersema secretly bought 50,000 copies of their ownbook. Made NY times best seller list. (Rational Herds, Chamely, 2004)

How do local decisions affect global decision making?Example: Sequential change detection has multiple threshold policy

decision

posterior probability of change

declare change

no changepolicy

Page 20: SPTM Expert Panel - IEEE Signal Processing Societysignalprocessingsociety.org/uploads/Publications/SPM/... · 2018-12-21 · SPTM Expert Panel In Network Processing, Adaptation, and

Sensor-Adaptive Signal Processing – Economics viewpoint

Noise

Signal

SensorSignal Processing

Estimate

Feedback (Stochastic Control)

Topic 3: Global Game for

Sensor Activation

! Gives a communication free

deployment protocol.

! Simple threshold policies are in Nash

equilibrium.

! If uncertainty is high or congestion is low.

! Compare with other approaches:

! Local communication & self-organization

[Biswas & Phoba 06, Clare & Pottie 99,

Aroraa et. al. 04]

! Flocking

!"#$%&'$()*"+(),-.

/*)&

!-#,01#&

How can agents autonomously manage their behavior?Game theory can be used as a synthesis and analysis tool. How cansimple behavior by agents result in sophisticated global behavior? How toachieve consensus in decisions? (S. Hart, Econmetrica, 2005).

How can agents learn from actions of other agents?Social learning results in herding behavior.In 1995, Treacy & Wiersema secretly bought 50,000 copies of their ownbook. Made NY times best seller list. (Rational Herds, Chamely, 2004)

How do local decisions affect global decision making?Example: Sequential change detection has multiple threshold policy

decision

posterior probability of change

declare change

no changepolicy