An Evolutionary Field Theorem: Evolutionary Field Optimization in Training of Power-Weighted Multiplicative Neurons...

Area of Science:

• Evolutionary Computation
• Machine Learning
• Artificial Neural Networks

Background:

• Neuroevolutionary machine learning integrates evolutionary computation with neural networks for data-driven engineering.
• Existing methods may lack efficiency in complex, nonlinear modeling tasks.

Purpose of the Study:

• To introduce an evolutionary field theorem and a novel algorithm, Evolutionary Field Optimization with Geometric Strategies (EFO-GS).
• To develop modified Power-Weighted Multiplicative (PWM) neural models capable of representing polynomial nonlinearity.
• To apply EFO-GS for training PWM models for efficient neuroevolutionary computation.

Main Methods:

• Development of an evolutionary field theorem and the EFO-GS algorithm featuring field-adapted differential crossover and metamutation.
• Modification of multiplicative neuron models to create PWM units, enabling real-valued, complex-valued, and mixed-mode operations.
• Training of PWM neural models using the EFO-GS algorithm for neuroevolutionary computation.

Main Results:

• The EFO-GS algorithm demonstrates improved evolutionary search quality through specialized mechanisms.
• PWM neural models effectively represent polynomial nonlinearity and operate in multiple modes.
• The combined EFO-GS and PWM approach achieved accurate estimation of Nitrogen Oxides (NOx) concentrations in an electronic nose application.

Conclusions:

• The proposed EFO-GS algorithm and PWM neural models offer a powerful framework for neuroevolutionary machine learning.
• This approach enhances the performance of data-driven engineering applications, particularly in sensor-based pollutant estimation.
• The study highlights the potential of these advancements for practical, real-world applications.