Artificial evolution by viability rather than competition.

Area of Science:

• Computational intelligence
• Artificial evolution
• Optimization algorithms

Background:

• Evolutionary algorithms (EAs) are heuristic methods for complex problems.
• Discovering diverse solutions alongside optimal ones is often desired.
• Defining a single fitness function for multiple objectives is challenging and can limit solution diversity.

Purpose of the Study:

• To present an alternative abstraction of artificial evolution that does not require a composite fitness function.
• To improve the discovery of diverse solutions in evolutionary computation.

Main Methods:

• Inspired by viability theory, natural evolution, and ethology.
• Focuses on eliminating individuals not meeting changing viability criteria (objectives and constraints).
• Avoids traditional competition-based selection and composite fitness functions.

Main Results:

• The proposed method maintains higher population diversity compared to classical EAs.
• It generates a greater number of unique solutions.
• Experimental results demonstrate improved performance over competition-based methods.

Conclusions:

• Incorporating viability principles into evolutionary algorithms enhances their applicability and effectiveness.
• This approach offers a powerful alternative for complex optimization problems in science and engineering.
• The method shows promise for applications like protein structure prediction and aircraft wing design.