Evolution and Learning Braitenberg 6: Selection, the Impersonal Engineer Braitenberg 7: Concepts EXAMPLES: Modeling evolution: genetic algorithms Modeling learning: reinforcement learning (bee foraging)
Genetic Algorithms genetic coding (chromosome/genes) genotype phenotype (development) evaluation, of individuals in a population selection, based on fitness variation, genetic modification of selected repeat
sensorL sensorR Parameters to encode: Input gains: 2 Time constants: 2 Thresholds: 4 Injection current: 4 Injection noise: 4 Synaptic weights: 4 Syn. time constants: 4 Output gains: 2 TOTAL: 26 Assume 4 bit resolution per parameter (16 possible values): 26 * 4 = 104 bits 2104 possibilities !!! motorL motorR
Learning vs. evolution
Functional organization (bee brain)
Reinforcement Learning Montague PR, Dayan P, Person C, Sejnowski TJ (1995) Bee foraging in uncertain environments using predictive Hebbian learning. Nature 377, 725 - 728 S R P N B r(t) Sensory Input Nectar Action δ(t) WB WN WY Y Diagram from: http://www.cs.stir.ac.uk/~kjt/teaching/31z7/posters/2003/igb.ppt