16
Swarm Intelligence: Where Biology meets Computers Tomas Piatrik Multimedia and Vision Research Group Queen Mary University of London Multimedia and Vision Research Group

Swarm Intelligence: Where Biology meets Computers · Swarm Intelligence: Where Biology meets Computers Tomas Piatrik Multimedia and Vision Research Group Queen Mary University of

  • Upload
    others

  • View
    7

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Swarm Intelligence: Where Biology meets Computers · Swarm Intelligence: Where Biology meets Computers Tomas Piatrik Multimedia and Vision Research Group Queen Mary University of

Swarm Intelligence: Where Biology meets Computers

Tomas Piatrik

Multimedia and Vision Research Group

Queen Mary University of London

Multimedia and Vision Research Group

Page 2: Swarm Intelligence: Where Biology meets Computers · Swarm Intelligence: Where Biology meets Computers Tomas Piatrik Multimedia and Vision Research Group Queen Mary University of

Overview

• Nature and Biology in Computer Science

• Popular Biological Inspired Systems

• Swarm Intelligence

– Ant Colony Optimisation

• Implementation of ACO for Clustering of Visual Data

• Conclusions

Multimedia and Vision Research Group

Page 3: Swarm Intelligence: Where Biology meets Computers · Swarm Intelligence: Where Biology meets Computers Tomas Piatrik Multimedia and Vision Research Group Queen Mary University of

Nature and Biology in Computing

• The Power of Nature – Connectionism

– Social Behaviour

– Emergence

– Adaptation

• Systems inspired by nature: – Evolutionary algorithms -- River formation dynamics --

Intelligent Water Drops -- Gravitational Search -- Stochastic Diffusion Search -- Particle Swarm Optimisation -- Ant Colony Optimisation -- Artificial Neural Networks -- Artificial Immune Systems . . . etc.

Multimedia and Vision Research Group

Page 4: Swarm Intelligence: Where Biology meets Computers · Swarm Intelligence: Where Biology meets Computers Tomas Piatrik Multimedia and Vision Research Group Queen Mary University of

Popular Biologically Inspired Systems

• Genetic algorithms – inheritance, – mutation, – natural selection, – genetic crossover

• Artificial Neural Networks

– neurons – connections – weights – local processing, distributed memory, synaptic weight dynamics and

weight modification by experience

• Artificial Immune Systems – learning and memory – clonal selection, negative selection, immune networks

Multimedia and Vision Research Group

Page 5: Swarm Intelligence: Where Biology meets Computers · Swarm Intelligence: Where Biology meets Computers Tomas Piatrik Multimedia and Vision Research Group Queen Mary University of

Swarm Intelligence

• What is it?

Characteristics:

• simple agents

• little knowledge

• decentralised control

• self organisation

• global behaviour

• able to solve complex problems

Multimedia and Vision Research Group

Page 6: Swarm Intelligence: Where Biology meets Computers · Swarm Intelligence: Where Biology meets Computers Tomas Piatrik Multimedia and Vision Research Group Queen Mary University of

Ant Colony Optimisation

• Meta-Heuristic

• Pheromone

• Shortest Path

• The probability of choosing direction:

Multimedia and Vision Research Group

otherwise , 0

)())((

)())((

)(allowedk

ijij

ijij

k

ijt

t

tp

ijijij tt )1()(

m

k

k

ijij

1

0

,k

k

ij L

Q

Pheromone value:

Page 7: Swarm Intelligence: Where Biology meets Computers · Swarm Intelligence: Where Biology meets Computers Tomas Piatrik Multimedia and Vision Research Group Queen Mary University of

Image Clustering using ACO

• The probability of clustering image i to cluster j:

• Updating pheromone level for all images:

Multimedia and Vision Research Group

K

j

CXCX

CXCX

CX

jiji

jiji

jip

0

),(),(

),(),(

),(

m

a

CXCXCX tttjijiji

1

),(),(),( )()1()(

0

)(

),(

0

ij

),(

K

i

i

ji

CX CsumDist

CCDistN

ji

),(),(

ji

CXCXDist

Bji

Page 8: Swarm Intelligence: Where Biology meets Computers · Swarm Intelligence: Where Biology meets Computers Tomas Piatrik Multimedia and Vision Research Group Queen Mary University of

Ant Tree Strategy for Visual Classification

• Inspired by self-assembling behavior of African ants and their ability to build chains (bridges) by their bodies in order to link leaves together.

• We model the ability of ants to build live structures with their bodies in order to discover, in a distributed and unsupervised way, a tree-structured organization of the visual data.

Multimedia and Vision Research Group

Page 9: Swarm Intelligence: Where Biology meets Computers · Swarm Intelligence: Where Biology meets Computers Tomas Piatrik Multimedia and Vision Research Group Queen Mary University of

• General principles

each ant: represents node of tree (data)

- ao support, apos position of moving ant

- incoming links; other ants maintain

toward ant i

Ant-Tree Strategy

• Main algorithm 1. all ants placed on the support;

initialization: Tsim(i)=1, Tdissim(i)=0

2. While there exists non connected ant i Do

3. If ant i is located on the support Then Support case

4. Else Ant case

5. End While

Multimedia and Vision Research Group

Page 10: Swarm Intelligence: Where Biology meets Computers · Swarm Intelligence: Where Biology meets Computers Tomas Piatrik Multimedia and Vision Research Group Queen Mary University of

Proposed ACO for Feature Selection

Lion Building Rural Car Elephant Clouds

CSD, CLD

CLD, EHD TGF, EHD

CSD, TGF

CSD, DCD, GLC CSD, CLD

Multimedia and Vision Research Group Multimedia and Vision Research Group

Page 11: Swarm Intelligence: Where Biology meets Computers · Swarm Intelligence: Where Biology meets Computers Tomas Piatrik Multimedia and Vision Research Group Queen Mary University of

• The Corel image database - 600 images with 6 semantic concepts

• The Window on the UK 2000” database - 390 images with 6 sets

Experimental Evaluation

Lions Rural Buildings Cars Elephants Clouds

Boats Fields Vehicles Trees Buildings Roads

Page 12: Swarm Intelligence: Where Biology meets Computers · Swarm Intelligence: Where Biology meets Computers Tomas Piatrik Multimedia and Vision Research Group Queen Mary University of

• The Caltech Image Database - 3550 images, 40 semantic concepts

Watch Water Lilly Wild cat Scissors Yin Yang

Ant Pizza Face Dollar Bill Snoopy

Soccer ball Stop Sign Strawberry Sunflower Umbrella

Experimental Evaluation

Multimedia and Vision Research Group

Page 13: Swarm Intelligence: Where Biology meets Computers · Swarm Intelligence: Where Biology meets Computers Tomas Piatrik Multimedia and Vision Research Group Queen Mary University of

• Flickr Image Database - 500 images segmented into regions, 4320 regions

• Semantic Concepts: Sand, Sea, Vegetation, Building, Sky, Person, Rock, Tree, Grass, Ground, and.

Experimental Evaluation

Multimedia and Vision Research Group

Page 14: Swarm Intelligence: Where Biology meets Computers · Swarm Intelligence: Where Biology meets Computers Tomas Piatrik Multimedia and Vision Research Group Queen Mary University of

Experimental Evaluation

Page 15: Swarm Intelligence: Where Biology meets Computers · Swarm Intelligence: Where Biology meets Computers Tomas Piatrik Multimedia and Vision Research Group Queen Mary University of

Original Camera video: 5min. Video summary: 53 sec.

• We cluster video frames according to similarity of the content. • From each cluster, one video segment is taken as a part of the summary

Set of representative frames for each scene:

Ant-tree algorithm

• Inspired by self-assembling behavior of African ants.

• We model the ability of ants to build live structures with their bodies in order to discover, in a distributed and unsupervised way, a tree-structured organization and summarization of the video data.

Frames are placed to the tree according to visual similarity and temporal information

Video Summarisation using Ants

Page 16: Swarm Intelligence: Where Biology meets Computers · Swarm Intelligence: Where Biology meets Computers Tomas Piatrik Multimedia and Vision Research Group Queen Mary University of

Conclusion

Multimedia and Vision Research Group

• Nature and Biology – Source of inspiration for scientists

• Importance of building artifacts whose inspiration derives from biology and the natural world

THANK YOU