Transcript On Artificial Chemistries

EvoDebate
Statement
September 2001
Wolfgang Banzhaf
Universität Dortmund, Informatik XI
Overview
• Extending the paradigm
– Methods or Approaches
– Theory
– Applications
08.05.2001
W. Banzhaf - LS XI, FB Informatik,
Universität Dortmund
2
Extending the Paradigm
• Evolutionary Computation
– Simplest approaches were introduced first
– Simplest models were studied first
– Toy problems were applied first
• Possible routes to extension
– Riding the Technology wave
– Problems in Science etc. are demanding
– AL/AC, MC, QC, ER, ...
08.05.2001
W. Banzhaf - LS XI, FB Informatik,
Universität Dortmund
3
EC - From Simple to Complicated
• Methods
–
–
–
–
GA: From bits to everything
ES: From real numbers to various representations
EP: From FSA to other representations
GP: From trees to all kinds of structures
• Theory
– Schema Inequality (GA) to Schema equation
– Spherical mutation rates (ES) to mutation ellipsoid
• Application
– Regression Problem (GP) to classification and prediction
– Parameter Optimization (GA/ES) to Multi-objective Opt.
08.05.2001
W. Banzhaf - LS XI, FB Informatik,
Universität Dortmund
4
Where to go from here?
• Riding the technology wave
– Moore‘s Law
– Distributed Networks
– Scientific Progress
• Problems are demanding
– Automatic Development of Scientific Models
– Design Process for all Sorts of Devices
– Complexity of Control
• New Arrivals
–
–
–
–
AL – AC
MC
QC
ER
08.05.2001
W. Banzhaf - LS XI, FB Informatik,
Universität Dortmund
5
Breeding of Algorithms and Machines
Possible to breed (presently)
•
Simple programs (GP)
•
Electrical circuits (GP)
•
Molecular machines (proteins) (Evolution in-vitro)
•
Strains of bacteria (in-vitro, artificial selection)
•
Organs (in-vitro)
Not possible to breed presently
•
Robots (hardware)
•
Other real macroscopic machines
08.05.2001
W. Banzhaf - LS XI, FB Informatik,
Universität Dortmund
6
Problems
•
What representation to choose?
–
–
•
Substrate for evolution
Operators able to work on the substrate
How to reach universal formability of the underlying substrate (matter,
structures)?
–
–
•
Developmental process of biology
Self-assembly, Self-organization
How to code complex functions into „fitness“ for breeding?
–
–
08.05.2001
Explicit vs. implicit fitness
Pre-structuring (time-scale separation, e.g. evolutionary and adaptive)
W. Banzhaf - LS XI, FB Informatik,
Universität Dortmund
7