Evolutionary Computation
Download
Report
Transcript Evolutionary Computation
Natural Computation
and Its Applications
Xin Yao
Natural Computation Group
Email: [email protected]
http://www.cs.bham.ac.uk/~xin
What This Lecture is NOT About
Not Commercial
Not Programming
Not Even Lecturing!
Frustration with Computers
Brittle
Non-adaptive
Doesn’t learn
Hopeless in dealing with noisy and inaccurate
information
Doesn’t do the homework for me although I
told it that I want a mark over 70%
…
Mother Nature
Who designed us and all our wonderful capabilities?
Natural Computation
Nature-Inspired
Computation
Natural Computation
Evolutionary computation
Neural computation
Molecular computation
Quantum computation
Ecological computation
Biological computation
…
Evolutionary Algorithm: An Example
Initialise the population
Repeat until the halting criteria are met
–
–
–
Fitness evaluation
Parent selection (natural selection)
Breeding/reproduction by crossover and mutation to
generate the new generation
Comparison of Four Methods
http://www.evonet.polytechnique.fr/CIRCUS2/node.php
?node=71
Moving Target
http://www.evonet.polytechnique.fr/CIRCUS2/node.php
?node=73
Evolving a Nozzle
http://www.evonet.polytechnique.fr/CIRCUS2/node.php
?node=72
Ant Colony Optimisation
Channel Allocation Inspired by Fruit Flies
Fruitflies have an insensitive exoskeleton peppered with sensors formed
from short bristles attached to nerve cells. It is important that the
bristles are more or less evenly spread out across the surface of the
fly. In particular it is undesirable to have two bristles right next to each
other. The correct pattern is formed during the fly's development by
interactions among its cells. The individual cells "argue" with each
other by secreting protein signals, and perceiving the signals of their
neighbours. The cells are autonomous, each running its own
"algorithm" using information from its local environment. Each cell
sends a signal to its neighbours; at the same time it listens for such a
signal from its neighbours. The signal is saying, in effect, "I want to
make a bristle". The more "loudly" it "hears" its neighbours signalling,
the less of the signal it produces. In other words the signal is
inhibitory. This "arguing" process is the inspiration for the channel
allocation method presented here.
Container Packing
How to pack a standard size container with
various sized boxes to minimise wasted
space?
How cut a standard length stock according to
different requirements while minimising
wastage?
…
Applications of Evolutionary
Computation
Genetic Algorithms in Parametric Design of Aircraft
Air-Injected Hydrocyclone Optimization Via Genetic
Algorithm
A Genetic Algorithm Approach to Multiple Fault Diagnosis
A Genetic Algorithm for Conformational Analysis of DNA
Automated Parameter Tuning for Sonar Information
Processing
http://www.nutechsolutions.com/case_studies/
Neural Computation
Parallel and distributed
Learnable
Fault-tolerant
Noise-tolerant
Efficient computation from slow components!
Good at perception tasks
…
Artificial Life
Life as it could be vs. life as it is
Great at exploring the huge space of artefacts
Boids
Karl Sims’s artificial creatures
…
Evolutionary Art
Evolutionary art from Andrew Rowbottom
Genetic art by Peter Kleiweg
Organic art by William Latham
By our own student!
…
Where to Find More information
MSc in Natural Computation
The Natural Computation Group
CERCIA (The Centre of Excellence for
Research in Computational Intelligence and
Applications)
AI/NC Seminars
MSc in Natural Computation
EPSRC studentships available, covering tuition
fees and maintenance costs, great as a
stepping stone for a PhD
Lots of industrial partners, good for a company
career
Small class size with lots of interactions with
lecturers
Programme Structure
TERM 1
1. Neural Computation
2. Evolutionary
Computation
TERM2
1. Nature-inspired
Optimisation
2. Nature-inspired
Learning
3. Quantum and
Molecular
Computation
3. Nature-inspired
Design
4. Mini-project 1
4. Mini-project 2
TERM3
Main project (some in
collaboration with an
industrial partner)
Natural Computation Group
One of the strongest in the world
7 core academic members and more than 20
PhD students
4 other teaching staff with strong overlaps
CERCIA
Four research fellows (additional to NC group
staff) and three admin staff
Specialise in applied research and industrial
projects
Current work includes energy consumption
prediction, evolutionary art, business match,
etc.
Summary
Ever-increasing complexity of the problems to
be solved by computers and the everincreasing complexity of the computer systems
require a radical rethinking of future directions
of computing
Natural computation (nature inspired
computation) is a promising future direction