Evolving neural networks through bio-inspired parent selection in dynamic environments.

Area of Science:

• Evolutionary Computation
• Artificial Intelligence
• Bio-inspired Algorithms

Background:

• Environmental variability challenges algorithm convergence in search spaces.
• Existing bio-inspired methods focus on crossover, mutation, and selection.
• Need for adaptive algorithms in dynamic environments.

Purpose of the Study:

• Propose NEAT-HD, a novel bio-inspired approach for evolutionary algorithms.
• Enhance algorithm performance and robustness in uncertain environments.
• Investigate the impact of genetic similarity-based parent selection.

Main Methods:

• Developed NEAT-HD, a bio-inspired algorithm.
• Incorporated a sigmoid function to accelerate species formation and population diversity.
• Utilized genetic similarity for parent selection.
• Conducted experiments on two classic control tasks.

Main Results:

• NEAT-HD demonstrated dynamic adaptation to environments through hybrid individuals from distinct parents.
• Observed a significant increase in population diversity via hybrid and novel individuals.
• Algorithm performance improved with optimized genetic distance distribution settings.
• Newly formed individuals exhibited robustness against dynamic environmental changes.

Conclusions:

• NEAT-HD offers a robust solution for evolutionary algorithms facing environmental uncertainty.
• The approach effectively enhances population diversity and promotes adaptation.
• Genetic similarity-based selection and sigmoid function are key to its success.