23
by Eric Han, Chung Min Kim, and Kathryn Tarver Investigations of Ant Colony Optimization

By Eric Han, Chung Min Kim, and Kathryn Tarver Investigations of Ant Colony Optimization

Embed Size (px)

DESCRIPTION

How Ants Forage for Food 1.Random walk 2.Pheromone is dropped 3.Food source quality affects pheromone amount 4.More pheromone = favored path 5.Pheromone evaporates

Citation preview

Page 1: By Eric Han, Chung Min Kim, and Kathryn Tarver Investigations of Ant Colony Optimization

by Eric Han, Chung Min Kim, and Kathryn Tarver

Investigations of Ant Colony Optimization

Page 2: By Eric Han, Chung Min Kim, and Kathryn Tarver Investigations of Ant Colony Optimization

An Introduction to Ants

10,000+ species of ants around the world

Eat seeds, nectar, fungi, insects, etc.

Colonies led by queens

Page 3: By Eric Han, Chung Min Kim, and Kathryn Tarver Investigations of Ant Colony Optimization

How Ants Forage for Food

1. Random walk

2. Pheromone is dropped

3. Food source quality affects pheromone amount

4. More pheromone = favored path

5. Pheromone evaporates

Page 4: By Eric Han, Chung Min Kim, and Kathryn Tarver Investigations of Ant Colony Optimization

ACO in Action

Page 5: By Eric Han, Chung Min Kim, and Kathryn Tarver Investigations of Ant Colony Optimization

ACO in Action

Page 6: By Eric Han, Chung Min Kim, and Kathryn Tarver Investigations of Ant Colony Optimization

ACO in Action

Page 7: By Eric Han, Chung Min Kim, and Kathryn Tarver Investigations of Ant Colony Optimization

ACO in Action

Page 8: By Eric Han, Chung Min Kim, and Kathryn Tarver Investigations of Ant Colony Optimization

ACO: Ant Colony Optimization

● First suggested by Marco Dorigo (1992)

● Inspired by foraging ant colonies

● Algorithm sends particles on random walks to optimize

pathways

● Currently applied to problems such as Internet routing

and protein folding

Page 9: By Eric Han, Chung Min Kim, and Kathryn Tarver Investigations of Ant Colony Optimization

Our goal is to:

1. create an algorithm to find the shortest path between

two points in a network, and

2. explore the effects of changing parameters in the

algorithm.

Project Goal

Page 10: By Eric Han, Chung Min Kim, and Kathryn Tarver Investigations of Ant Colony Optimization

Pseudocode

for each iteration:

1. run ants

2. add pheromone

3. evaporate pheromone

Page 11: By Eric Han, Chung Min Kim, and Kathryn Tarver Investigations of Ant Colony Optimization

Objects

0

1

23

4

2

2

1

53

1

Page 12: By Eric Han, Chung Min Kim, and Kathryn Tarver Investigations of Ant Colony Optimization

Parameters

Parameter Description

p Rate of pheromone evaporation

q Scalar, proportional to amount of pheromone added to path

colsize Number of ants

Page 13: By Eric Han, Chung Min Kim, and Kathryn Tarver Investigations of Ant Colony Optimization

Pseudocode: The Antsrun antsfor each ant: while not at end vertex: mark current vertex as visited for all unvisited vertices: roll RNG to see if traveling this vertex if traveling: move to the vertex add vertex to path

Page 14: By Eric Han, Chung Min Kim, and Kathryn Tarver Investigations of Ant Colony Optimization

Pseudocode: The Edgesadd pheromone: for each ant for each edge along its path add (Q÷L) pheromone to the edge

evaporate pheromone: for each edge multiply pheromone value by (1-p)

Page 16: By Eric Han, Chung Min Kim, and Kathryn Tarver Investigations of Ant Colony Optimization

Observation 1More ants → Less Noise, fewer convergences onto local optimal

Page 17: By Eric Han, Chung Min Kim, and Kathryn Tarver Investigations of Ant Colony Optimization

Observation 2

Weight scaling → decreases # ants taking optimal path

Page 18: By Eric Han, Chung Min Kim, and Kathryn Tarver Investigations of Ant Colony Optimization

Observation 3

Increasing pheromone evaporation rate → increases % of ants taking optimal path

Page 19: By Eric Han, Chung Min Kim, and Kathryn Tarver Investigations of Ant Colony Optimization

Observation 4

Increasing amount of pheromone added → no effect

Page 20: By Eric Han, Chung Min Kim, and Kathryn Tarver Investigations of Ant Colony Optimization

General Observations● Large, dense graph

o sometimes will find global optimalo usually will converge on local optimal

comes close to the global optimalo need many ants to avoid local optimal

● Small, dense graph o ants almost always find global optimalo don’t need as many ants or iterations to do soo converges more slowly

Page 21: By Eric Han, Chung Min Kim, and Kathryn Tarver Investigations of Ant Colony Optimization

Conclusions

● To maximize ants taking best path:

o high evaporation rate

o large colony

o smaller path weights

Page 22: By Eric Han, Chung Min Kim, and Kathryn Tarver Investigations of Ant Colony Optimization

Discussion: What Now?● Dynamic graphs● Eliminate convergences onto local optimal● Optimize running time● Analytically determine effects of changing parameters

Page 23: By Eric Han, Chung Min Kim, and Kathryn Tarver Investigations of Ant Colony Optimization

BibliographyAnts, Ant Pictures, Ant Facts - National Geographic. (n.d.). Retrieved July 20, 2015.

Argentine Ant l Globe spanning insect society - Our Breathing Planet. (n.d.). Retrieved July 20, 2015.

Blum, C., & Li, X. (2008). Swarm Intelligence in Optimization. Natural Computing Series Swarm Intelligence, 43-85. Retrieved July 20, 2015.

Priyadi, A. Ant fire [Online image]. Retrieved July 20, 2015 from http://yourshot.nationalgeographic.com/photos/3098725/?source=gallery.

Ant clipart [Online image]. (2014). Retrieved July 30, 2015 from ……… http://www.clipartpanda.com/clipart_images/ant-clipart-158500

http://www.pageresource.com/clipart/clipart/animals/insects/ants/ant-3.png

Thank you!