Organic Evolution and Problem Solving

1
Organic Evolution and Problem Solving

• Je-Gun Joung

2
1.2 Evolutionary Algorithms and Artificial
Intelligence

• A definition of artificial Intelligence by Rich
• Artificial Intelligence is the study of how to
  make computers do things at which, at the moment,
  people are better.
• Some researchers in AI propose to orientate
  toward imitation of the much more restricted
  capabilities of less complex animals

3
Representation

• The symbolic period of AI can be dated period
  from 1962 until 1975.
• A knowledge-intensive period from 1976 until
  1988.
• Currently, the field of AI is starting to spread
  research into a variety of directions
• Subsymbolic period of AI dates from 1950 until
  1965.
• Evolutionary Algorithms make use of a subsymbolic
  representation of knowledge encoded in the
  genotypes of individuals

4
Learning Characteristic

• Rote learning No inference processes take place.
  Instead, direct implantation of knowledge is
  performed.
• Learning by instruction This term denotes
  knowledge acquisition from a teacher or from an
  organized source and integration with existing
  knowledge.
• Learning by deduction Deductive,
  truth-preserving inferences and memorization of
  useful conclusions are summarized by this term

5
Learning Characteristic (2)

• Learning by analogy The transformation of
  existing knowledge that bears strong similarity
  to the desired new concept into a form
  effectively useful in the new situation
• Learning by induction Inductive inferences
• Learning from examples (concept acquisition)
• Learning by observation and discovery
  (descriptive generalization, unsupervised
  learning)

6
Artificial Life

• Artificial Life research concentrates on computer
  simulations of simple hypothetical life forms
• The problem how to make their behavior adaptive.
• Self-organizing properties emerging from local
  interactions within a large number of simple
  basic agents are investigated.
• Analogies to natural systems can be drawn on a
  variety of different levels.
• In many cases the agents are equipped with
  internal rules of strategies determining their
  behavior

7
1.4 Early Approaches

• Attempts to model natural evolution as a method
  for searching for good solutions of problems
  defined on vast search spaces.
• Very restricted computer power was available at
  that time
• Automatic programming, sequence prediction,
  numerical optimization, and optimal control

8
Automatic Programming

• Finding a program which calculate a certain
  input-output function
• An attempt towards evolving computer programs as
  performed by Friedberg et al. In 1958
• Binary encoded
• Modification by instruction interchange and
  random changes of instructions
• Success number for instructions
• The mutation rate depended on the success numbers

9
Automatic Programming (2)

• Selective pressure
• To test different programs created by random
  instruction changes and instruction interchanges
• To choose the best of the new programs as the
  next starting point.
• The approach measured quality of the program by
  combining the binary feedback information

10
Optimization

• Bremermanns work was more oriented towards
  optimization in 1962.
• Multiple mutations are necessary to overcome
  points of stagnation
• Optimal mutation probability-1/l

11
Evolutionary Programming

• Forgel 1964
• A more complicated application domain
• A sequence prediction problem (
  finite-state-machine FSM)
• Population-based algorithm

12
Evolutionary Operation (EVOP)

• EVOP approach as presented by Box in 1957.
• This method emphasized on the natural model of
  the organic evolution by performing a
  mutation-selection process.
• ( ) -strategy (where or
  , the so-called 22 and 23 factorial design method

13
1.5 Summary

• The basic process of transcription and
  translation , the genetic code and the
  hierarchical structure of genetic information
• In connection to meiotic heredity, the crossover
  mechanism and the various forms of mutation
• Evolution processes on the lower level of
  biological macromolecules

14
Summary

• Evolutionary algorithms are inductive learning
  algorithms that can serve as a powerful search
  method in many fields of AI research.
• Three examples of global optimization problems
• Computational complexity of global optimization
  problems
• The early approaches