Evolutionary optimization and neural network models of behavior.

Area of Science:

• Behavioral Ecology
• Evolutionary Ecology
• Computational Neuroscience

Background:

• Organisms face constraints from ontogeny and phylogeny, limiting their ability to achieve theoretically optimal behaviors.
• Adaptationist programs and optimization models in ecology struggle with these biological limitations.

Purpose of the Study:

• To compare the predictive power of optimality models versus neural network models for organismal behavior.
• To investigate if neural networks can achieve near-optimal fitness despite evolutionary and developmental constraints.

Main Methods:

• Developed an evolutionary optimization model to calculate behaviors maximizing Darwinian fitness for polychaete worms.
• Constructed a neural network model incorporating motor, sensory, energetic, and clock neuronal groups, with adaptable connections.
• Simulated neural network ontogeny (individual learning) and evolution (natural selection of parameters).

Main Results:

• The best neural network models achieved 85% to 99% of the fitness predicted by the evolutionary optimization model for polychaete worms.
• Neural network models for tephritid fruit flies generated predictions regarding their host acceptance neurobiology.
• Demonstrated that neural networks are sufficiently complex and plastic for organisms to attain near-optimal fitness.

Conclusions:

• Neural networks provide a powerful framework for understanding how organisms achieve high fitness despite constraints.
• Even minor evolution of neural network 'design parameters' can lead to near-optimal behaviors, especially for simpler actions.
• This approach bridges evolutionary theory with neurobiological mechanisms, offering insights into the evolution of behavior.