Interactive evolutionary computation with minimum fitness evaluation requirement and offline algorithm design.

Area of Science:

• Artificial Intelligence
• Computational Intelligence
• Human-Computer Interaction

Background:

• Interactive Evolutionary Computation (IEC) relies on human feedback for solution evaluation.
• Traditional IEC models often require extensive human evaluation, which is impractical in many real-world applications like hearing aid design.
• Human evaluations can be subjective and imprecise, necessitating models with minimal evaluation demands.

Purpose of the Study:

• To formulate an IEC model that minimizes the fitness evaluation ability required from human users.
• To develop and evaluate an algorithm tailored for this low-requirement IEC model.
• To introduce an automated algorithm design approach for optimizing IEC processes.

Main Methods:

• A novel IEC model is proposed, assuming users can evaluate only one solution at a time and remember only the immediately preceding one.
• A (1+1)ES-style algorithm is developed for the proposed IEC model, focusing on efficient solution generation.
• An offline meta-level approach is introduced for automated algorithm design, utilizing varied solution generation mechanisms (mutation, crossover, random initialization).

Main Results:

• Computational experiments demonstrate the effectiveness of the proposed approach.
• The novel IEC model and its associated algorithms show competitive or superior performance compared to traditional methods.
• The automated algorithm design approach, using diverse generation mechanisms, outperforms fixed-mechanism algorithms.

Conclusions:

• The proposed IEC model effectively addresses the limitations of human evaluation in complex design tasks.
• The developed (1+1)ES-style algorithm and the meta-level automated design approach offer efficient solutions for IEC.
• This research advances the field of IEC by enabling effective optimization with minimal human input.