Multi-objective cooperative coevolution of artificial neural networks (multi-objective cooperative networks)

Area of Science:

• Artificial Intelligence
• Computational Neuroscience
• Machine Learning

Background:

• Evolving neural networks traditionally focuses on whole network structures.
• Assigning credit to individual components within evolving networks is challenging.
• Cooperative coevolutionary models require effective methods for multi-criteria fitness evaluation.

Purpose of the Study:

• To introduce MOBNET, a cooperative coevolutionary model for evolving neural network topology and weights.
• To address the credit assignment problem using multi-objective optimization for subcomponents.
• To demonstrate the competitive performance and efficiency of MOBNET.

Main Methods:

• Developed MOBNET, a model that evolves neural network subcomponents.
• Applied multi-objective optimization to evaluate the fitness of subcomponents based on conflicting criteria.
• Compared MOBNET against standard classification methods and other neural network models.

Main Results:

• MOBNET demonstrated highly competitive performance across four real-world problems.
• The model achieved superior overall performance compared to all other classification methods applied.
• MOBNET generated smaller neural networks than existing models.

Conclusions:

• Multi-objective optimization is an effective approach for evaluating subcomponent fitness in cooperative coevolution.
• MOBNET offers a promising new direction for neural network evolution and coevolutionary computation.
• The MOBNET framework is extensible to broader applications of coevolutionary algorithms.