decentralised coordination of continuously valued control parameters using the max-sum algorithm...

44
Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick Jennings School of Electronics and Computer Science University of Southampton {rs06r, af2, acr, nrj}@ecs.soton.ac.uk

Post on 15-Jan-2016

221 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

Decentralised Coordination of Continuously Valued Control Parameters using the

Max-Sum Algorithm 

Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick JenningsSchool of Electronics and Computer Science

University of Southampton{rs06r, af2, acr, nrj}@ecs.soton.ac.uk

Page 2: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

2

This presentation focuses on the use of Max-Sum in coordination problems with continuous parameters

From Discrete to Continuous

Max-Sum for Decentralised Coordination

Empirical Evaluation

Page 3: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

3

Max-Sum is a powerful algorithm for solving DCOPs

Complete Algorithms

DPOPOptAPOADOPT

Communication Cost

Iterative Algorithms

Best Response (BR)Distributed Stochastic

Algorithm (DSA) Fictitious Play (FP)

Max-SumAlgorithm

Optimality

Page 4: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

Max-Sum solves the social welfare maximisation problem in a decentralised way

Agents

Page 5: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

Max-Sum solves the social welfare maximisation problem in a decentralised way

1x

2x

3x

4x

5x

6x

7x8x

Control Parameters

Page 6: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

Max-Sum solves the social welfare maximisation problem in a decentralised way

1U

2U

3U

4U

5U

6U

7U8U

Utility Functions

Page 7: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

Max-Sum solves the social welfare maximisation problem in a decentralised way

)( 33 xU

Localised Interaction

},,,{ 54313 xxxxx

Page 8: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

Max-Sum solves the social welfare maximisation problem in a decentralised way

Agents

M

iiiU

1

)(maxarg xx

Social welfare:

Page 9: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

The input for the Max-Sum algorithm is a graphical representation of the problem: a Factor Graph

Variable nodes Function nodes

1x

2x

3x

1U

2U

3U

Agent 1Agent 2

Agent 3

Page 10: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

Max-Sum solves the social welfare maximisation problem by message passing

1x

2x

3x

1U

2U

3U

Variable nodes Function nodes

Agent 1Agent 2

Agent 3

Page 11: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

Max-Sum solves the social welfare maximisation problem by message passing

jiadjk

iikiji xrxq\)(

)()(

ijadjk

kjkjji

iij xqUxrj \)(\

)()(max)( xx

From variable i to function j

From function j to variable i

Page 12: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

Until now, Max-Sum was only defined for discretely valued variables

Graph Colouring)( iji xq )( iij xr

Page 13: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

However, many problems are inherently continuous. Heading

andVelocity

Unattended Ground Sensor

ActivationTime

Autonomous Ground Robot

Thermostat

Preferred RoomTemperature

Page 14: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

So, we extended the Max-Sum algorithm to operate in continuous action spaces

Discrete Continuous

Page 15: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

We focussed on utility functions that are Continuous Piecewise Linear Functions (CPLFs)

Page 16: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

We focussed on utility functions that are Continuous Piecewise Linear Functions (CPLFs)

)( iji xq )( iij xr

“Continuous” Graph Colouring

Page 17: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

A CPLF is defined by a domain partitioning followed by value assignment

Page 18: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

A CPLF is defined by a domain partitioning followed by value assignment

Page 19: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

A CPLF is defined by a domain partitioning followed by value assignment

Page 20: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

ijadjk

kjkjji

iij xqUxrj \)(\

)()(max)( xx

To make Max-Sum work on CPLFs, we need to define key two operations on them

jiadjk

iikiji xrxq\)(

)()(

From variable i to function j

From function j to variable i

Page 21: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

ijadjk

kjkjji

iij xqUxrj \)(\

)()(max)( xx

To make Max-Sum work on CPLFs, we need to define key two operations on them

jiadjk

iikiji xrxq\)(

)()(

From variable i to function j

From function j to variable i1. Addition

of two CPLFs

Page 22: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

ijadjk

kjkjji

iij xqUxrj \)(\

)()(max)( xx

To make Max-Sum work on CPLFs, we need to define key two operations on them

jiadjk

iikiji xrxq\)(

)()(

From variable i to function j

From function j to variable i

2. Marginal Maximisation to a single variable

Page 23: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

Addition of two CPLFs involves merging their domains, and then summing their values

1x

2x

1x

2x

1f 2f

Page 24: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

Addition of two CPLFs involves merging their domains, and then summing their values

1x

2x

1x

2x

1x

2x

1. Merge domains

1f 2f

21 ff

Page 25: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

Addition of two CPLFs involves merging their domains, and then summing their values

Page 26: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

Addition of two CPLFs involves merging their domains, and then summing their values

2. Sum Values

Page 27: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

Marginal maximisation is the operation of finding the maximum value of a function, if we fix all but one variable

ijadjk

kjkjji

iij xqUxrj \)(\

)()(max)( xx

From function j to variable i:

Page 28: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

Marginal maximisation involves finding the maximum value of a function, if we fix all but one variable

ijadjk

kjkjji

iij xqUxrj \)(\

)()(max)( xx

)(max)(\

jji

i fxgj

xx

Page 29: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

Marginal maximisation involves finding the maximum value of a function, if we fix all but one variable

ijadjk

kjkjji

iij xqUxrj \)(\

)()(max)( xx

)(max)(\

ji

i fxgj

xx

),(max)( 2112

xxfxgx

Example: bivariate function:

Page 30: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

Marginal maximisation involves the projection of a CLPF on a 2-D plane, and upper envelope extraction

),( 1 fxProject onto

axis

Page 31: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

Marginal maximisation involves the projection of a CLPF on a 2-D plane, and upper envelope extraction

Project onto

axis),( 1 fx

Page 32: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

Marginal maximisation involves the projection of a CLPF on a 2-D plane, and upper envelope extraction

Project onto

axis),( 1 fx

Result of projection

Page 33: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

Marginal maximisation involves the projection of a CLPF on a 2-D plane, and upper envelope extraction

Extract UpperEnvelope

Page 34: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

Marginal maximisation involves the projection of a CLPF on a 2-D plane, and upper envelope extraction

Extract UpperEnvelope

),(max)( 2112

xxfxgx

)( 1xg

Page 35: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

We empirically evaluated this algorithm in a wide-area surveillance scenario

Dense deployment of sensors to detect activity within an urban environment.

Unattended Ground

Sensor

Page 36: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

Sensors adapt their duty cycles to maximise event detection by coordinating with overlapping sensors

time

duty cycleDiscrete

time

duty cycle

time

duty cycle

Discretised time

Page 37: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

Sensors adapt their duty cycles to maximise event detection by coordinating with overlapping sensors

time

duty cycleDiscrete Continuous

time

duty cycle

time

duty cycle

time

duty cycle

time

duty cycle

time

duty cycle

Page 38: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

Continuous Max-Sum outperforms Discrete Max-Sum by up to 10%

0 5 10 15 20 25 300.75

0.80

0.85

0.90

0.95

1.00

ContinuousDiscrete

Discretisation

Solu

tion

Qua

lity

(as

frac

tion

of o

ptim

al)

Average Solution Quality over 25 Iterations

Page 39: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

Tota

l Mes

sage

Siz

eContinuous Max-Sum leads to more effective use of communication resources than Discrete Max-Sum

0 5 10 15 20 25 300

5000

10000

15000

20000

ContinuousDiscrete

Discretisation

Total number of values exchanged between agents

Page 40: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

In conclusion, we have shown that Continuous Max-Sum is more effective than Discrete Max-Sum

1. No artificial discretisation

time

time

Page 41: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

In conclusion, we have shown that Continuous Max-Sum is more effective than Discrete Max-Sum

1. No artificial discretisation Continuous

Discrete

2. Better solutions

time

timeSo

lutio

n Q

ualit

y

Page 42: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

In conclusion, we have shown that Continuous Max-Sum is more effective than Discrete Max-Sum

time

time1. No artificial discretisation Continuous

Discrete

2. Better solutions ContinuousDiscrete

3. Effective communication

Solu

tion

Qua

lity

Mes

sage

Siz

e

Page 43: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

For future work, we wish to extend the algorithm to arbitrary continuous functions

For example, using Gaussian Processes

Page 44: Decentralised Coordination of Continuously Valued Control Parameters using the Max-Sum Algorithm Ruben Stranders, Alessandro Farinelli, Alex Rogers, Nick

In conclusion, we have shown that Continuous Max-Sum is more effective than Discrete Max-Sum

time

time1. No artificial discretisation Continuous

Discrete

2. Better solutions ContinuousDiscrete

3. Effective communication

Solu

tion

Qua

lity

Mes

sage

Siz

eQuestions?