Genetic Algorithms - PowerPoint PPT Presentation

Loading...

PPT – Genetic Algorithms PowerPoint presentation | free to download - id: 3c7811-NWE4N



Loading


The Adobe Flash plugin is needed to view this content

Get the plugin now

View by Category
About This Presentation
Title:

Genetic Algorithms

Description:

Genetic Algorithms Genetic Algorithms Genetic Algorithms Part II Sanchit Arora 2005CS10182 Group 1 Outline Overview of Presentation I Strengths and Limitations ... – PowerPoint PPT presentation

Number of Views:166
Avg rating:3.0/5.0
Slides: 27
Provided by: seasUpen2
Category:

less

Write a Comment
User Comments (0)
Transcript and Presenter's Notes

Title: Genetic Algorithms


1
Genetic Algorithms
Genetic Algorithms Genetic Algorithms Part II
  • Sanchit Arora
  • 2005CS10182
  • Group 1

2
Outline
  • Overview of Presentation I
  • Strengths and Limitations
  • Application Domains
  • Various Examples

3
Overview of Presentation I
  • Inspiration Evolutionary Biology
  • Selection
  • Inheritance
  • Mutation
  • Crossover (recombination)
  • Setup
  • Representation
  • Fitness Function

4
Flowchart of a Genetic Algorithm
5
Strengths Limitations
  • Intrinsically parallel
  • Well suited to problems with large solution space
  • Works even for complex fitness landscapes
  • Noisy
  • Discontinuous
  • Changes over time
  • Many local optima
  • Ability to manipulate many parameters
    simultaneously
  • Open-minded approach
  • Negative feedback
  • Defining problem representation
  • Writing fitness function
  • Other parameters
  • Size of population
  • Rate of mutation and crossover
  • Type and strength of selection
  • Deceptive fitness functions
  • Premature convergence
  • Unnecessary for analytically solvable problems

6
Application Domains
  • Electrical engineering
  • Financial markets
  • Game playing
  • Image Enhancement
  • Mathematics and algorithms
  • Molecular biology
  • Pattern recognition and data mining
  • Robotics
  • Routing and Scheduling
  • Systems Engineering

7
Robotics Applications
  • Using Genetic Algorithms for Robot Motion
    Planning
  • GOAL Build dynamic motion planners for robots
    with 6 degrees of freedom
  • Used Genetic algorithms for
  • Inverse Kinematics problem
  • Path planning problem

8
Why Genetic Algorithms
  • Well adapted to search in high dimensionality
    space
  • Tolerant of function form
  • Implementation on massively parallel machines
  • Achieves super linear speed up with no of
    processors

9
Inverse kinematics problem
Direct Kinematics Function
Desired Cartesian location
10
Using GAs
  • Coding the problem
  • Discretized search space from 0,2px0,2p
  • I1,I2 ? 0,,255
  • b I1I2 -gt (0000000100000011) binary
  • Generating initial population
  • n Random individuals
  • Operating a selection
  • f applied to each individual
  • Ranked on basis of f value
  • Creating couples and combining individuals
  • bi, bj randomly chosen on basis of rank
  • Cross-over used to produce new individuals
  • Generate mutants (optional)

11
A simple path planner
  • Simple path planner for a planar arm with two
    degrees of freedom
  • Search space
  • Discretized subset of all the possible paths
  • Manhattan motion M(k,q)
  • Length of path k
  • Bits of discrete ? q

12
GA setup
  • Coding search space element
  • P ? M(k,q) -gt k x q bits
  • Defining fitness function

13
Results
  • Scans intermediary points of candidate path to
    find a direct move to the goal
  • Stops when obstacle is found and return distance
    from extremity of last free segment
  • Results Only around 1/10 of the configuration
    space is actually evaluated

14
Image Enhancement
  • A survey of Genetic Algorithms Applications for
    Image Enhancement and Segmentation
  • Image segmentation using edge detection
  • Character recognition

15
Image Segmentation using Edge Detection
  • Two stages
  • Edge enhancement
  • Derivative evaluation
  • Threshold
  • Selection and combination of edge map pixels
  • Boundary detection
  • Edge linking
  • Grouping of local edges

16
GA based search for optimal configuration of
pixels
  • Possible edge configuration S
  • Encoded as chromosome K2 bit string
  • Each bit presence of edge pixel
  • Cost Function
  • Fragmentation
  • Thickness
  • Local length
  • Region similarity
  • Curvature
  • Evaluated for each pixel in M x M neighborhood

17
Cost analysis
  • Fragmentation
  • Local edge discontinuities
  • Penalty endpoints of the edge
  • endpoint only 1 neighbor or isolated
  • Thickness
  • Edge strength
  • Penalty Thin edges
  • thinness only 1 connectivity

18
Cost analysis cont..
  • Length
  • To avoid excessive edges
  • Penalty assigned to each pixel
  • Removes noise and short/useless edge fragments
  • Region Similarity
  • Estimating likelihood of edge
  • Penalty proportional to dissimilarity
  • Curvature/Smoothness
  • Pixel-to-Pixel connection angle
  • Penalty proportional to angle

19
Other Steps
  • Initial edge configurations
  • Generated from filtered image
  • Because of large search space (2k2)
  • Reproduction / Mutation
  • Trivial
  • Termination
  • When cost function becomes invariant over
    generations

20
Extension of previous algorithm
  • Each chromosome encodes only small portion of
    image as a 8x8 window
  • Connectivity between windows maintained
  • Edge connectivity at corners
  • Problem based mutation operation
  • Selects a mutation strategy from a set of
    predefined mutations
  • Decreased convergence time

21
Character recognition
  • Most difficult
  • Separating character from background
  • Complex variations
  • Lighting conditions
  • Variety of degradations
  • Difficult for a single filtering technique to
    deal with a variety of degradations
  • GAs -gt construct an optimal sequence of image
    processing filters to extract characters from
    different sources

22
Setup
  • Degradations
  • Clear
  • Background with pattern
  • Character with pattern
  • Character with rims
  • Blurring
  • Non-uniform lighting
  • A filter bank of 17 well-known filters
  • mean, min, max, sobel, erosion, dilation etc
  • Search for optimal filtering sequence

23
Fitness function
  • Fitness f(T,F)
  • Compared filtered image with image ideally
    segmented by human
  • Kx width Ky height

24
Details
  • Each chromosome string of 8 bit integers
    representing filter index
  • Total length limited to 80
  • Initial population 300 individuals
  • Max generations 800
  • Fitness threshold 0.9
  • Selection Roulette wheel selection

25
References
  • Wikipedia
  • The TalkOrigins Archive
  • http//www.talkorigins.org/faqs/genalg/genalg.html
  • Using Genetic Algorithms for Robot Motion
    Planning
  • http//citeseerx.ist.psu.edu/viewdoc/summary?doi1
    0.1.1.54.4217
  • A survey of Genetic Algorithms Applications for
    Image Enhancement and Segmentation
  • http//itc.ktu.lt/itc363/Paulinas363.pdf

26
Thank You
  • The random chance of variation, coupled with the
    law of selection, is a problem-solving technique
    of immense power and nearly unlimited
    application
  • an insight on results by Charles Darwin
About PowerShow.com