Interactive Evolutionary Computation

1
Interactive Evolutionary Computation Two initial
definitions of IEC (Takagi, 2001) 1) EC
optimises the target systems based upon
subjective human evaluation as fitness functions
for systems outputs (narrow definition) 2) EC
optimises the target systems via an interactive
human-machine interface(broad definition). Intent
ion of the Workshop is to explore these
definitions through practical examples both
broadening them whilst introducing finer-grained
categorisation.
2

• Narrow Definition
• User represents a black box evaluation
  function
• User operates within a psychological space
  rather than a parametric space e.g. evaluates
  system output (e.g. sound, graphics etc) rather
  than parametric (phenotype) description.
• Takes into account fluctuating human
  preferences that are generally impossible to
  model.
• Techniques identify regions of interest
  containing a number of preferred solutions (close
  in psychological space but possibly diverse in
  parametric space?).
• Human fatigue (cognitive overload?) factor
  restricts number of generations.

3

• Broad Definition
• Tends to relate to Evolutionary Multi-objective
  satisfaction where human evaluation of objective
  preferences is required along with qualitative
  judgement of quantitative solutions.
• Also relates to human input into evolutionary
  process in terms of introduction of elite
  solutions / radical solutions, variation of
  constraints, addition / subtraction of variables.

4

• More Detailed Definition?
• Three categories may be more appropriate
• Subjective selection - where interaction relates
  to human selection of favorite option along with,
  perhaps, a simple ranking of available solutions
  
• Subjective evaluation - where a more rigorous
  human evaluation of available solutions takes
  place, possible machine-learning component.
• Problem definition - where user interacts with
  evolutionary system to gather optimal information
  which supports the modification / reformulation
  of the evaluation function. Agent-based support
  and machine-learning.

5
Workshop Format Presentation 1 Origins of
Interactive Evolutionary Computation - Graphic
Art, Music and Fashion I. C. Parmee Short
Discussion Period Presentation 2 Applications
of Interactive Evolutionary Computation
P. Caleb-Solly Short Discussion
Period Invitation to the floor - short
presentations, extended comments Break for
Coffee
6
Presentation 3 Adaptive Image Segmentation
Based on Visual Interactive Feedback
Learning P. Caleb-Solly Short
Discussion Period Presentation 4
Poor-definition, Uncertainty and Human Factors
- A Case for Interactive Evolutionary Problem
Reformulation? I. C. Parmee Short
Discussion Period Invitation to the
floor Round- table Session