Neural Network Evolving Algorithm Based on the Triplet Codon Encoding Method.

Area of Science:

• Artificial Intelligence
• Computational Neuroscience
• Bioinformatics

Background:

• Neural Evolution (NE) is a key area in Artificial Intelligence (AI) that uses evolutionary algorithms to design Artificial Neural Networks (ANNs).
• A significant challenge in NE is optimizing both network topology and weights for practical applications.
• Existing NE methods face limitations in effectively evolving complex network structures and parameters.

Purpose of the Study:

• To propose a novel DNA encoding method for Neural Evolution based on the triple codon structure.
• To introduce the Triplet Codon Encoding Neural Network Evolving Algorithm (TCENNE) to validate the proposed encoding scheme.
• To evaluate the effectiveness and robustness of TCENNE compared to existing NE algorithms.

Main Methods:

• Development of a new DNA encoding strategy utilizing triple codon principles.
• Implementation of the Triplet Codon Encoding Neural Network Evolving Algorithm (TCENNE).
• Testing TCENNE on challenging reinforcement learning tasks to assess performance.

Main Results:

• The proposed TCENNE algorithm demonstrates high effectiveness and robustness.
• TCENNE successfully enables the co-evolution of network topology and weights.
• The algorithm outperforms existing neural evolution systems in complex reinforcement learning scenarios.

Conclusions:

• The novel DNA encoding method based on triple codons provides a robust foundation for Neural Evolution.
• TCENNE offers a significant advancement in evolving Artificial Neural Networks, particularly for reinforcement learning.
• This approach facilitates the simultaneous optimization of network architecture and parameters, leading to superior performance.