Download Evolutionary Algorithms - Computer Network Lab.

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Hologenome theory of evolution wikipedia , lookup

Introduction to evolution wikipedia , lookup

Genetics and the Origin of Species wikipedia , lookup

Evolutionary psychology wikipedia , lookup

Catholic Church and evolution wikipedia , lookup

State switching wikipedia , lookup

Theistic evolution wikipedia , lookup

Saltation (biology) wikipedia , lookup

Darwinian literary studies wikipedia , lookup

Evolutionary landscape wikipedia , lookup

The eclipse of Darwinism wikipedia , lookup

Evolutionary mismatch wikipedia , lookup

Transcript
A New Tool to Cross the Hurdles:
Evolutionary Algorithms
April 12, 2012
Prof. Chang Wook Ahn
Sungkyunkwan Evolutionary Algorithms Lab.
Dept. of Computer Engineerng
Sungkyunkwan University
SEAL.
Contents
l Principle of Evolutionary Algorithms
Ø
Ø
Ø
Ø
Prologue
Principle
Conventional Approach
Operational Concept
l Real-World Application
Ø
Ø
Ø
Ø
Ø
Ø
Ø
Combinatorial Problems
Communication & Networks
Economic Science
Game, Art, Music, Design
Information Mining, Artificial Creatures
Recommender Systems
Swarm Robotics
Principle of
Evolutionary
Algorithms
SEAL.
Prologue (1)
l Where are Evolutionary Algorithms (EAs) placed?
Problem Solving
Techniques
Heuristic
Approach
Deterministic
Approach
Simplex Method
Nature-Inspired
Methods
Other Methods
Genetic Algorithms
Random Search
Linear Programming
Particle Swarm Opt.
Bayesian Inference
Gradient Descent
Ant Colony Opt.
SEAL.
Prologue (2)
l Where to be Applied?
Problems
elbavlosnU
Hilbert’s 10th Problem
Turing’s Halt Problem
………
Tractable:
- Solve the problems in a polynomial time;
O(nk), not O(n!) or O(2n)
P: Polynomial
NP: Nondeterministic Polynomial
NP-hard
Untractable
elbavloS
Hamiltonian Path
Longest Path
………
NP
NPC
NP-Complete
P
Tractable
Shortest Path Problem
Minimum Spanning Tree
………
If a problem turns out to be NPC/NP-hard,
Stop to find a deterministic algorithm!
In this case, EAs become a good tool!
SEAL.
Principle (1)
l What are EAs ?
Ø An algorithmic abstraction inspired from the theory of biological evolution,
usually implemented on computers, which is employed for resolving problems
Key Components of
Biological Evolution
Natural Selection
Genetic Inheritance
Survival of
the fittest!
There is much
similarity!
SEAL.
Principle (2)
l Lessons from Biological Evolution
Implications for applying
to computing techs.
Multiple
Surviving
POPULATION
MATING POOL
MATES SELECTED
MATING
Mixing
OFFSPRING
Generation
NEW POPULATION
SEAL.
Conventional Approach
l What’s the Problem of Conventional (Search) Approaches?
Single nodal case
Optimum
Multiple nodal case
Suboptimum
SEAL.
Operational Concept
l Main Principle of EAs
Ø Multiple individuals try to cooperatively resolve problems
by mimicking evolutionary mechanisms
Surviving
Multiple
Mixing
Generation
Suboptima
Optimum
Real-World Applications
SEAL.
Classical Combinatorial Problems
Traveling Salesman Problem
Knapsack Problem
- Maximize the amount of profits (e.g., money)
while still keeping the overall weight
under or equal to a given limit!
SEAL.
Communication & Networks
Multicast Routing
- Minimize the cost of multicast tree
while satisfying delay and bandwidth
constraints
Resource Allocation
- Maximize resource utilization
by fairly distributing wireless resources
among the connections
SEAL.
Economic Science
Time-Series Forecasting
Decision in Dilemma
Deny?
Confess?
34
32
30
28
26
24
22
20
18
5
10
15
20
25
30
35
40
45
50
- Predicting some future outcomes from
a set of historical events
- Stock prediction, Weather forecasting, etc.
- Choosing a decision in conflict objectives
- Prisoner’s dilemma, Game theory, etc.
SEAL.
Game
l Evolutionary Checker
l Video Game: NERO
Ø 8X8 board, 12 checkers for each player
- Diagonal moves, Jumps are forced, etc.
Ø Neural Networks + Evolutionary Prog.
- Checkerboards are evaluated by NNs
- NNs and King value are evolved with EP
Ø Almost the expert level without knowledge
Ø Univ. of Texas at Austin
Ø Player’s role
- Train agents for competition
Ø No prepackaged or scripted agents
Ø Evolve in real-time
SEAL.
Art
l What’s Evolutionary Art?
Ø Technically, it is creating pieces of art through human-computer interaction
Ø Computer runs evolutionary algorithms and human applies subjective selection
- Role of computers: offer choices and create diversity
- Role of human: make (subjective) choices and reduce diversity
Ø Selection (aesthetic/subjective) steers towards implicit user preferences
Evol. Art by Kleiweg
Galapagos by Karl Sims
Other Examples
SEAL.
Music
l GenJam (Genetic Jammer)
Source:
http://www.it.rit.edu/~jab/
GenJam.html
Ø
Ø
Ø
Ø
Developed in 1993~94 by Prof. John Al Biles
Interactive GA that leans to play jazz solos
GemJam’s repertoire: Over 250 jazz-style tunes
Evolving by special fitness operators;
e.g., rhythm conformity
Ø What can it be done?
ü Playing full-chorus improvised solos
ü Listening to trumpet and responds
interactively when we trade fours
ü Engaging in collective improvisation;
we both solo simultaneously and GenJam
performs a smart echo of improvisation
ü Listening to me and play the head of a tune
and breeds my measures
Source: http://phoenix.inf.upol.cz/~dostal/evm.html
Virtual
quintet
MusiGenesis
SEAL.
Design
l Structure Design
Ø Bridge structure optimization
Ø Building structure design
l Aviation System Design
Ø Airfoil, wing, and antenna designs
Ø Space platform structure optimization
Ø Jet aircraft model optimization
SEAL.
Information Mining
l Clustering
Ø Data clustering
Ø Text mining
Ø Web search
l Bioinformatics
Ø Drug discovery
Ø Protein folding
Ø Cancer diagnosis
SEAL.
Artificial Creatures & Robotics
l Artificial Creatures
Ø eFly, Biomorph, HAL,
ØSelf-replicating Worms
Ø Gozilla, Solitaire
l Robotics
Ø Humanoid Robots; e.g., e.g., ASIMO
Ø Genetic Robots; e.g., Gene
Ø Others; e.g., Six-Legged Robot
Robot Snake
SEAL.
Recommender Systems
v E-magazine Recommender System
ü Automatically recommend a set of magazines according to the user’s history
ü User’s behavior is learned by Interactive Evolutionary Computation
E-magazine System
Recommendation Module
Us
His er
tor
y
User
Initial
Magazine Set
Content Search
Social Search
Database
Recommendation
Module
=
Feature Extraction
Data Grouping
Recommend
New magazines
Content-filtering
User magazine set
Genetic Operator
SEAL.
Swarm Robotics (1)
v Nature-Inspired Foraging Swarm Robots
ü Improving Energy Efficiency in Foraging Swarm Robots Using Honey Bee Model
Foraging Robots Adopting Honey Bee Colony
Food Sources
Employed
foragers (workers)
Scouts
Onlookers
SEAL.
Swarm Robotics (2)
v Self-Assembling Swarm Robots
ü The system are built from modules, a kind of robotic cell
ü Each module is a robot having all the on-board components for creating a robot
Supervisor
Morphosis
Optimization
Motor Primitives
Robot Morphology
Environment
Morphosis controller is to decide/control when/how to change morphological properties.
Supervisor is to monitor performance, to switch between motor primitives for different gaits, to determine
when to run the optimization algorithms, and to provide a high-level interface for the human operator.
THANK YOU!
http://www.evolution.re.kr
[email protected]
Please take a close look around.
‘Evolution’ is happening all the time and everywhere!