An Evolutionary Field Theorem

This study contributes to the emerging field of neuroevolutionary machine learning in evolutionary computation. Firstly, it introduces an Evolutionary Field Optimization with Geometric Strategies (EFO-GS) algorithm based on an evolutionary field theorem, incorporating a field-adapted differential crossover and a field-aware metamutation process for improved search quality. Secondly, the study modifies the multiplicative neuron model to create Power-Weighted Multiplicative (PWM) neural models, capable of representing polynomial nonlinearity and operating in real-valued, complex-valued, and mixed modes. The EFO-GS algorithm is then employed to efficiently train the PWM neuron models. In an electronic nose application, this integrated approach accurately estimates Nitrogen Oxides (NOx) pollutant concentrations from low-cost multi-sensor array measurements, demonstrating enhanced estimation performance.

Paper Url: https://www.mdpi.com/1424-8220/22/10/3836