BITS Pilani

Pilani|Dubai|Goa|Hyderabad
Integration of NSGA-II with ANN
Mid-Sem Presentation
for
BITS C 331-Computer Projects

B. Aditya
2010A4PS155H
BITS Pilani
Pilani | Dubai | Goa | Hyderabad
Under the guidance of
Dr. Amit Kumar Gupta
By
B. Aditya
2010A4PS155H
BITS Pilani
Pilani | Dubai | Goa | Hyderabad
Multi-objective optimization
BITS Pilani, Deemed to be University under Section 3 of UGC Act, 1956



Multiobjective optimization is the process of
simultaneously optimizing two or more conflicting
objectives subject to certain constraints.

Multiobjective Optimization
3/27/2012 4
Examples of multi-objective optimization problems:-
Maximizing profit and minimizing the cost of a product.
Maximizing performance and minimizing fuel consumption of a vehicle.
Minimizing weight while maximizing the strength of a particular
component.
BITS Pilani, Deemed to be University under Section 3 of UGC Act, 1956



Difference
3/27/2012 5
Single Objective Optimization
Optimize only one objective
function
Single optimal solution
Maximum/Minimum fitness value
is selected as the best solution.
Multiobjective Optimization
Optimize two or more than two
objective functions
Set of optimal solutions
Comparison of solutions by
Domination
Non-domination
Minimize



where -10 < x < 20

Optimal solution:-

0 10 20 30 40 50 60 70 80 90 100
0
10
20
30
40
50
60
70
80
90
100
f1
f
2

-10 -5 0 5 10 15 20
0
100
200
300
400
x

f
(
x
)



f1(x)
f2(x)
BITS Pilani, Deemed to be University under Section 3 of UGC Act, 1956
3/27/2012 6
Standard Approach :Weighted Sum
of Objective Functions
Limitations:
Result depends on weights.
Some solutions may be missed.
Multiple runs of the algorithm are required in order to get the whole
range of solutions.
Difficult to select proper combination of weights.
Combining objectives loses information and predetermines trade-offs
between objectives.

) ( ) ( ) ( ) (
2 2 1 1
x f x f x f x g
m m

+ + + =
{ } ) ( , ), ( ), (
2 ` 1
x f x f x f Minimize
m

1 &
2 1
= + + +
m

m
, ,. ,
2 1
where are weights values
and m represents the number of objective functions.
Formulate as a single objective with weighted sum of all objective functions -
BITS Pilani, Deemed to be University under Section 3 of UGC Act, 1956
Interactive methods
Interactive Surrogate Worth Trade-Off Method
Geoffrion-Dyer-Feinberg Method
Tchebycheff Method
Reference Point Method
GUESS Method
Satisficing Trade-Off Method
Light Beam Search
NIMBUS Method
No-preference methods
Meth. of Global Criterion
A posteriori methods
Weighting Method
c-Constraint Method
Hybrid Method
Method of Weighted Metrics
Achievement Scalarizing Function Approach
A priori methods
Value Function Method
Lexicographic Ordering
Goal Programming
Methods
BITS Pilani, Deemed to be University under Section 3 of UGC Act, 1956
Definitions
Domination :
One solution is said to dominate
another if it is better in all objectives.

Non-Domination [Pareto points] :
A solution is said to be non-
dominated if it is better than other
solutions in at least one objective.
3/27/2012 8
Minimize function
M
i
n
i
m
i
z
e

f
u
n
c
t
i
o
n

A
B
C
D
1
f
2
f
A dominates B (better in both and )
A dominates C (same in but better in )
A does not dominate D (non-dominated points)
A and D are in the Pareto optimal front
These non-dominated solutions are called Pareto optimal solutions.
This non-dominated curve is said to be Pareto front.
2
f
1
f
2
f
1
f
BITS Pilani, Deemed to be University under Section 3 of UGC Act, 1956
3/27/2012 9
Definitions
Pareto Optimal
A vector variable is Pareto optimal if for every
and either

or, there is at least one such that

where is the vector of decision variables,
is the vector of objective
functions, is the feasible region ,where represents
the whole search space.
BITS Pilani, Deemed to be University under Section 3 of UGC Act, 1956
Definitions Cont.
Pareto Optimal Set
For a given MOP the Pareto optimal set is
defined as

Pareto Front
For a given MOP and Pareto optimal set , the
Pareto front is defined as


3/27/2012 10
BITS Pilani, Deemed to be University under Section 3 of UGC Act, 1956
Many different approaches
VEGA, RWGA, MOGA, NSGA II, DPGA, etc.
Goals: maintaining diversity and guaranteeing Pareto
optimality how to measure?
Special operators have been introduced, fitness evaluated
in many different ways etc.
Problem: with real problems, it remains unknown how far
the solutions generated are from the true PO solutions
Multiobjective Evolutionary Algorithms
BITS Pilani, Deemed to be University under Section 3 of UGC Act, 1956
12 3/27/2012
Desirable MOEA features
Best Solutions

Lie on true Pareto front

They are uniformly
distributed on the front
Aim:
To achieve convergence to Pareto optimal front

To achieve diversity (representation of the entire Pareto
optimal front)
Minimize function
M
i
n
i
m
i
z
e

f
u
n
c
t
i
o
n

1
f
2
f
True Pareto
front
Possible
solutions
BITS Pilani, Deemed to be University under Section 3 of UGC Act, 1956
Performance Measures
3/27/2012 13
There are two main goals in a multi-objective optimization:
1) Convergence to the Pareto-optimal set
2) Maintenance of diversity in solutions of the Pareto-
optimal set.
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
0
2
4
6
8
10
12
function - 1
f
u
n
c
t
i
o
n

-

2
DISTANCE MEASURE BETWEEN PARETO FRONTS


TRUE PARETO FRONT
PARETO FRONT
min distance
min distance
Convergence metric
BITS Pilani, Deemed to be University under Section 3 of UGC Act, 1956
Comparison of MOGA Methods
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
0 1 2 3 4 5
f
1
f
2
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
0 1 2 3 4 5 f
1
f
2
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
0 1 2 3 4
f
1
f
2
Objective Functions :
f
1
(x) = x
2
; f
2
(x) = (x-2)
2

Figure 1. VEGA Results
Figure 2. NSGA Results
Figure 3. NSGA II Results
Figure 1
Figure 3
Figure 2
BITS Pilani
Pilani | Dubai | Goa | Hyderabad
NSGA-II- Fast and Elitist Genetic
Algorithm.
BITS Pilani, Deemed to be University under Section 3 of UGC Act, 1956
Report Final
Population and Stop
Begin : Initialize
Population (N)
Evaluate objective
functions
Non-dominated
Sorting
Tournament Selection
Combine parent and
child populations ,
Non-dominating
Sorting
Crossover &
Mutation
Evaluate objective
functions
Flowchart of NSGA-II
Yes
No
Select N
individuals
3/27/2012 16
Is max.
gens.
reached?
BITS Pilani, Deemed to be University under Section 3 of UGC Act, 1956
The population is initialized based on the problem range
and constraints if any.
A random number is picked between the minimum and
maximum possible values for the each decision variable.
This function initializes the chromosomes. Each
chromosome has the following at this stage:
set of decision variables
objective function values
Population intialization
BITS Pilani, Deemed to be University under Section 3 of UGC Act, 1956
Function to evaluate the objective functions for the given
input vector.
The algorithm always minimizes the objective function
hence if you would like to maximize the function then
multiply the function by negative one.

Evaluate objective
BITS Pilani, Deemed to be University under Section 3 of UGC Act, 1956
The initialized
population is
sorted based on
non-domination.
The fast sort
algorithm is
described as
below.

Non-Domination Sort
BITS Pilani, Deemed to be University under Section 3 of UGC Act, 1956
1
1
1
1
2
2
3
4
5
6
-0.414 0.171 5.829
0.467 0.218 2.347
0.818 0.669 1.396
1.735 3.011 0.07
3.210 10.308 1.465
-1.272 1.618 10.708
-1.508 2.275 12.308
-1.832 3.355 14.682
-2.161 4.671 17.317
-4.105 16.854 37.275
x ( ) x f
1
( ) x f
2
Rank
Fast Non-domination Sorting
Assigning the rank to each individual of the population.
Rank based on the non-domination sorting (front wise).
It helps in selection and sorting.
2 } , , , {
1 } , , {
1 } , {
0 } {
0 } {
0 } {
0 } {
7 6 3 2 1 7
6 3 2 1 6
5 4 3 5
4 4
3 3
2 2
1 1
= =
= =
= =
= =
= =
= =
= =
n x x x x x
n x x x x
n x x x
n x
n x
n x
n x

Reference
chromosome
Dominated
chromosomes
Counter
3/27/2012 20
BITS Pilani, Deemed to be University under Section 3 of UGC Act, 1956
3/27/2012 21
-0.414 0.171 5.829 1
0.467 0.218 2.347 1
0.818 0.669 1.396 1
1.735 3.011 0.07 1
3.210 10.308 1.465 2
-1.272 1.618 10.708 2
-1.508 2.275 12.308 3
-1.832 3.355 14.682 4
-2.161 4.671 17.317 5
-4.105 16.854 37.275 6
x ( ) x f
1
( ) x f
2
Rank
0 2 4 6 8 10 12 14 16 18
0
5
10
15
20
25
30
35
40
f
1
(x)
f
2
(
x
)
4
3
5
6
2
1
Fast Non-domination Sorting
BITS Pilani, Deemed to be University under Section 3 of UGC Act, 1956
This algorithm is better than the original NSGA ( [5])
since it utilize the information about the set that an
individual dominate (Sp) and number of individuals that
dominate the individual (np).

Non-domination Sort (contd.)
BITS Pilani, Deemed to be University under Section 3 of UGC Act, 1956
Crowding Distance Assignment
3/27/2012
23
To get an estimate of density of
solutions surrounding a particular
solution in population.

Choose individuals having large
crowding distance.

Help for obtaining uniformly
distribution.
where represent objective function value of solution.

and is the maximum value of function in the Pareto front.
max
m
f
m
f

) 1 ( , , 3 , 2 = l i
where

, 1
. . . .
= =
D C D C
l

|
|
.
|

\
|

+
=
m
m m
m m
D C
f f
i f i f
i
min max
. .
] 1 [ ] 1 [
BITS Pilani, Deemed to be University under Section 3 of UGC Act, 1956
Selection
3/27/2012 24
Selection is the stage of a genetic algorithm in which individual are chosen from a
population for later breeding (recombination or crossover).
The crowding operator guides the selection process at the various stages
of the algorithm toward a uniformly spread-out Pareto optimal front.
where shows non-domination rank & is crowding
distance of individual.
Crowding operator based sorting
j i
n
s
( )
rank rank
j i <
n
s
. .D C
i
rank
i
th
i
( ) ( )
. . . . D C D C rank rank
j i and j i
or
> =
BITS Pilani, Deemed to be University under Section 3 of UGC Act, 1956
3/27/2012 25
Runs a tournament among a few individuals chosen at random from the
population and selects the winner (the one with the best fitness) for crossover.

In tournament selection, a number Tour size of individuals is chosen
randomly from the population and the best individual from this group is
selected as parent. (Based on the crowding operator)
Tournament Selection

0.818 0.669 1.396 1 1.378
-1.508 2.275 12.30 3
x ( ) x f
1
( ) x f
2
Rank . .D C

0.818 0.669 1.396 1 1.378
rank rank
2 1 <

0.467 0.218 2.347 1 0.945
0.818 0.669 1.396 1 1.378
x ( ) x f
1
( ) x f
2
Rank . .D C

0.818 0.669 1.396 1 1.378
rank rank
2 1 =
. . . .
2 1
D C D C
<
BITS Pilani, Deemed to be University under Section 3 of UGC Act, 1956
where
is a random number {0,1}
is a crossover operator
represent dimension of
individual.
r

j
Crossover
3/27/2012 26
Crossover is a genetic operator that combines (mates) two individuals (parents)
to produce two new individuals (Childs).
The idea behind crossover is that the new chromosome may be better
than both of the parents if it takes the best characteristics from each of the
parents.
( )
( )

>
|
|
.
|

\
|

s
=
|
|
.
|

\
|
+
|
|
.
|

\
|
+
5 . 0
1 * 2
1
5 . 0 * 2
1
1
1
1
r if
r
r if r
b

Simulated Binary Crossover

( ) ( ) ( )
( ) ( ) ( ) ) ( * 1 ) ( * 1
2
1
) (
) ( * 1 ) ( * 1
2
1
) (
2 1 2
2 1 1
j parent b j parent b j child
j parent b j parent b j child
+ + =
+ + =
BITS Pilani, Deemed to be University under Section 3 of UGC Act, 1956
3/27/2012 27
Mutation
Mutation is a genetic operator that alters one ore more gene values in a
chromosome from its initial state.
Mutation is an important part of the genetic search as helps to prevent the
population from stagnating at any local optima.
Polynomial Mutation
( )
( ) ( )

>
s
=
|
|
.
|

\
|
+
|
|
.
|

\
|
+
5 . 0 1 * 2 1
5 . 0 1 * 2
1
1
1
1
r if r
r if r
d
q
q
d j parent j child + = ) ( ) (
where
is a random number {0,1}
is a mutation operator
represent dimension of
individual.
r
q
j
BITS Pilani, Deemed to be University under Section 3 of UGC Act, 1956
Pt
Qt
Rt ={Pt , Qt}

F1
Non-dominated
sorting (Rank)
Crowding distance
sorting
F2
F3
Rejected
Selection for next generation
Pt+1
Elitist Replacement
3/27/2012 28
BITS Pilani, Deemed to be University under Section 3 of UGC Act, 1956
Report Final
Population and Stop
Begin : Initialize
Population (N)
Evaluate objective
functions
Non-dominated
Sorting
Tournament Selection
Combine parent and
child populations ,
Non-dominating
Sorting
Crossover &
Mutation
Evaluate objective
functions
Flowchart of NSGA-II
Yes
No
Select N
individuals
3/27/2012 29
Is max.
gens.
reached?
BITS Pilani, Deemed to be University under Section 3 of UGC Act, 1956
References
3/27/2012 30
1. K. Deb, A. Pratap, S. Agarwal, and T. Meyarivan, A fast and elitist multi-objective genetic
algorithm: NSGA-II, IEEE Transaction on Evolutionary Computation, 6(2), 181-197,2002.

2. K. Deb and R. B. Agrawal, Simulated binary crossover for continuous search space, in
Complex Syst., vol. 9, pp. 115148., Apr. 1995.

3. N. Srinivas and K. Deb, Multiobjective function optimization using nondominated sorting
genetic algorithms, Evol. Comput., vol. 2, no. 3, pp. 221248, Fall 1995.

4. J. Horn, N. Nafploitis, and D. E. Goldberg, A niched Pareto genetic algorithm for
multiobjective optimization, in Proceedings of the First IEEE Conference on Evolutionary
Computation, Z. Michalewicz, Ed. Piscataway, NJ: IEEE Press, pp. 8287 , 1994.

5. J. D. Knowles and D.W. Corne, Approximating the nondominated front using the Pareto
archived evolution strategy, Evol. Comput., vol. 8, pp. 149172, 2000.

6. Carlos A. Coello Coello, Member, IEEE, Gregorio Toscano Pulido, and Maximino Salazar
Lechuga, Handling multiple objectives with particle swarm optimization, Evol. Comput.,
vol. 8, pp. 256279, No. 3, June 2004




BITS Pilani
Pilani|Dubai|Goa|Hyderabad
Thank you very much!