Introduction and motivation Revision Problems in graph theory Conclusion and references

Combinatorial optimization using Hopeld Net

Chintan B. Shah (06305013) Nirav S. Uchat (06305906)
Computer Science and Engineering IIT Bombay

Chintan B. Shah (06305013) Nirav S. Uchat (06305906)

Combinatorial optimization using Hopeld Net

Introduction and motivation Revision Problems in graph theory Conclusion and references

Introduction and Motivation

Optimization Problem
What is an Optimization Problem ? Many of them are NP-Complete (Hard in some sense) Finding an optimum solution is tough Finding a good solution may be easy

Why Hopeld Nets

Optimization problem maximizes/minimizes some quantity Hopeld Net minimizes energy This motivates us to model (and possibly solve) the optimization problem using Hopeld Net

Chintan B. Shah (06305013) Nirav S. Uchat (06305906)

Combinatorial optimization using Hopeld Net

Introduction and motivation Revision Problems in graph theory Conclusion and references

How to solve the problem using Hopeld Net

Choose the representation such that output of the neurons corresponds to the solution of the problem Choose the energy function for the problem such that its minimum value gives the best solution to the problem Compare the energy equation of the problem E p to the energy equation of the Hopeld Net En Find the weight and threshold values

Chintan B. Shah (06305013) Nirav S. Uchat (06305906)

Combinatorial optimization using Hopeld Net

Introduction and motivation Revision Problems in graph theory Conclusion and references

Agenda

Revision
Travelling Salesman Problem Sorting

Problems in Graph Theory

Vertex Cover Independent Set Clique Matching

Chintan B. Shah (06305013) Nirav S. Uchat (06305906)

Combinatorial optimization using Hopeld Net

Introduction and motivation Revision Problems in graph theory Conclusion and references

Travelling Salesman Problem

Energy Equations
E1 = E2 = i j Ep = E 1 + E 2 En = i
A 2 1 2 i( xi ,

1)2 +

xi , 1)2

dij xi (xj ,+1 + xj ,1 )

wi ,j xi xj

xi i

Weight and threshold values

Row Weight wij = A Column Weight wij = A Cross Weight wij = ij = 2A
dij +dji 2

Chintan B. Shah (06305013) Nirav S. Uchat (06305906)

Combinatorial optimization using Hopeld Net

Introduction and motivation Revision Problems in graph theory Conclusion and references

Sorting

Representation
Create a neuron corresponding to i th value and j th position If the neuron res, i th value will be in j th position

Equations
A value can be in at most 1 position A position can accomodate at most 1 value 2 2 E1 = A i ( xi , 1) + ( i xi , 1) 2 Summation over value position should be minimized E2 = B i xi S i

Chintan B. Shah (06305013) Nirav S. Uchat (06305906)

Combinatorial optimization using Hopeld Net

Introduction and motivation Revision Problems in graph theory Conclusion and references

Ep and En
Ep =
A 2 En i( xi , i

1)2 +
j

xi , 1)2 + B i wi ,j xi xj i xi i
( i

xi

Si

Weight and threshold values

wij = A = 2A Si B

Chintan B. Shah (06305013) Nirav S. Uchat (06305906)

Combinatorial optimization using Hopeld Net

Introduction and motivation Revision Problems in graph theory Conclusion and references

Vertex Cover (VC)

Representation
Create a neuron for each vertex If neuron res, the vertex will be in the VC

Equations
Minimize the no. of vertices in the VC E1 = Cover each edge E2 =
Chintan B. Shah (06305013) Nirav S. Uchat (06305906)

vi

dij vi vj

Combinatorial optimization using Hopeld Net

Introduction and motivation Revision Problems in graph theory Conclusion and references

Simplied energy equation

Ep = A =A =A =A
i i i i

dij vi vj + B i vi d j ij (1 vi vj ) + B i vi j dij (1 vi vj + vi vj ) + B i vi d v v A d ( v + v ) + B j j ij i j i j ij i
j

vi + A

dij

Ep and En
Ep = A En =
i i j

dij vi vj A j wij xi xj

i i

dij (vi + vj ) + B

vi

xi i

Weight and threshold values

wij = 2Adij ij = 2A i dij B

Chintan B. Shah (06305013) Nirav S. Uchat (06305906)

Combinatorial optimization using Hopeld Net

Introduction and motivation Revision Problems in graph theory Conclusion and references

Independent set (IS)

Representation
Create a neuron for each vertex If neuron res, the vertex will be in the IS

Equations
Maximize the no. of vertices in the IS E1 = E2 =
Chintan B. Shah (06305013) Nirav S. Uchat (06305906)

vi vj )

There shouldnt be an edge between the vertices in the IS

i j eij (vi

Combinatorial optimization using Hopeld Net

Introduction and motivation Revision Problems in graph theory Conclusion and references

Simplied energy equation

Ep = A =A
i i j eij (vi

vj ) + B B
i

vi

j eij (vi vj )

vi + B

Ep and En
Ep = A En =
i i

B i vi j wij xi xj i xi i

j eij (vi vj )

Weight and threshold values

wij = Aeij i = B

Chintan B. Shah (06305013) Nirav S. Uchat (06305906)

Combinatorial optimization using Hopeld Net

Introduction and motivation Revision Problems in graph theory Conclusion and references

Clique

Representation
Create a neuron for each vertex If neuron res, the vertex will be in the Clique

Equations
Maximize the no. of vertices in the Clique E1 = E2 =
j (eij i

vi 1)(vi vj )

Conrm that there is an edge between every pair of vertices

i

Chintan B. Shah (06305013) Nirav S. Uchat (06305906)

Combinatorial optimization using Hopeld Net

Introduction and motivation Revision Problems in graph theory Conclusion and references

Simplied energy equation

Ep = A = A
i i j (eij j (eij

1)(vi vj ) + B
i

vi

1)(vi vj ) B

vi + B

Ep and En
Ep = A En =
i i

1)(vi vj ) B j wij xi xj i xi i

j (eij

vi

Weight and threshold values

wij = Aeij 1 i = B

Chintan B. Shah (06305013) Nirav S. Uchat (06305906)

Combinatorial optimization using Hopeld Net

Introduction and motivation Revision Problems in graph theory Conclusion and references

Matching

Representation
Create a neuron for each edge If neuron res, the edge will be in the Matching

Equations
Maximize the no. of edges in the Matching E1 = E2 =
Chintan B. Shah (06305013) Nirav S. Uchat (06305906)

ei

The edges in the Matching shouldnt be adjacent

i j

dij (ei ej )

Combinatorial optimization using Hopeld Net

Introduction and motivation Revision Problems in graph theory Conclusion and references

Simplied energy equation

Ep = A =A
i i j j

dij (ei ej ) + B
i ei

ei

dij (ei ej ) B

+B

Ep and En
Ep = A En =
i i

dij (ei ej ) B i ei j wij xi xj i xi i

j

Weight and threshold values

wij = Adij i = B

Chintan B. Shah (06305013) Nirav S. Uchat (06305906)

Combinatorial optimization using Hopeld Net

Introduction and motivation Revision Problems in graph theory Conclusion and references

Conclusion
mathematical formulation for the problems (requires ingenuity) intuitive values for weights and thresholds cute programs

Chintan B. Shah (06305013) Nirav S. Uchat (06305906)

Combinatorial optimization using Hopeld Net

Introduction and motivation Revision Problems in graph theory Conclusion and references

References
Xiaoming CHEN, Zheng TANG, An Hopeld Network Learning for Minimum Vertex Cover Problem some references, SICE Annual Conference in Sapporo, pp. 1150-1155, 2004. Jenn-Shiang Lai, Sy-Yen Kuo, Ing-Yi Chen, Neural networks for optimization problems in graph theory, Circuits and Systems, ISCAS 94, Vol.6, pp. 269-272, 1994. J. Ramanujam and P. Sadayappan, Optimization using neural networks, in Proceedings of the 2nd IEEE International Conference on Neural Networks, vol. 2, pp. 325-332, San Diego, CA, July 1988 Mark Atkins, Sorting by Hopeld Net,IJCNN, Vol.2, pp. 65-68, 1989.
Chintan B. Shah (06305013) Nirav S. Uchat (06305906) Combinatorial optimization using Hopeld Net