Learning internal representations by error propagation
Parallel distributed processing: explorations in the microstructure of cognition, vol. 1
Adaptation in natural and artificial systems
Adaptation in natural and artificial systems
Genetic algorithms + data structures = evolution programs (2nd, extended ed.)
Genetic algorithms + data structures = evolution programs (2nd, extended ed.)
1994 Special Issue: Design and evolution of modular neural network architectures
Neural Networks - Special issue: models of neurodynamics and behavior
Genetic Algorithms in Search, Optimization and Machine Learning
Genetic Algorithms in Search, Optimization and Machine Learning
Emergent oscillations in evolutionary simulations: Oscillating networks increase switching efficacy
Journal of Cognitive Neuroscience
| Hi-index | 0.00 |
The evolutionary justification by LeDoux (1996) for his dual-route model of fear processing was analyzed computationally by applying genetic algorithms to artificial neural networks. The evolution was simulated of a neural network controlling an agent that gathered food in an artificial world and that was occasionally menaced by a predator. Connections could not change in the agent's "lifetime," so there was no learning in the simulations. Only if the smells of food and predator were hard to distinguish and the fitness reflected time pressures in escaping from the predator did the type of dual processing postulated by LeDoux emerge in the surviving agents. Processing in the "quick and dirty" pathway of the fear system ensured avoidance of both predators and food, but a distinction between food and predator was made only in the long pathway. Elaborate processing inhibited the avoidance reaction and reversed it into an approach reaction to food, but strengthened the avoidance reaction to predators (and more finely tuned the direction of escape). It is suggested that "computational neuroethology" (Beer, 1990) may help constrain reasoning in evolutionary psychology, particularly when applied to specific neurobiological models, and in the future may even generate new hypotheses for cognitive neuroscience.