14-InteractiveEAs
Interactive Evolutionary
Algorithms
Motivation
Characteristics
Approaches
Interactive evolution as design vs. optimization
Motivation
- Interactive evolution (IE) : measure of a solution’s fitness is
provided by a human’s subjective judgment
- World is full of examples of humanoid intervention (pets, food
crops)
- Applications of IE Algorithms: capturing aesthetics in art and
design, personalization of artifacts such as medical devices
- Human’s judgment
- Advantage: insight and guidance
- Disadvantage: inconsistent, loss of attention
Characteristics
- The user becomes effectively part of the system (like in
agricultural breeding)
- Features that impact on the design of IEAs:
- Effect of time:
- Avoid lengthy evolution and focus on making rapid gains to fit in
with human needs
- Human decision takes longer than evaluation mathematical fitness
function
- Effect of context:
- Human expectations change in response to what evolution
produces
- Advantages of IEAs:
- Handling situations with no clear fitness function
- Improved search ability, increased exploration and diversity
Algorithmic Approaches (1/2)
- Interactive selection and population size
- Subjective selection
- Direct (choosing individuals for reproduction)
- Indirect (assigning fitness, sorting)
- Use of small population because:
- Limited number of solutions can be shown
- When ranking, the pair-wise comparisons grow rapidly
- Multi-objective EAs are used for problems with mixture of
quantitative and qualitative aspects
- Intervention in the variation process
- Implicit: periodically adjust the choice and parameterization of
variation operators, using the given score to control mutation
- Explicit: inspect promising solutions to adjust them by hand and
place them back in to the population (Lamarkian)
- Use of surrogate fitness function
- Approximate the decision a human would make
- Advantage: can use large populations
Interactive evolution as design
vs. optimisation
IE is related to evolutionary art and design