Explaining Adaptation in Genetic Algorithms with Uniform Crossover
summary
Proposes the hyper-climbing hypothesis: genetic algorithms with uniform crossover implement a decimation-style global search that detects coarse schema partitions and fixes high-fitness loci. Uses a simple GA with independent per-bit mutation, sigma scaling and stochastic universal sampling, constructs staircase fitness functions and a sparse-parity proof-of-concept that recovers four parity loci among 10,000 irrelevant loci in about 500 generations. Claims polylogarithmic scaling in loci, and reports validations on Max3SAT and spin-glass instances.
AI-written summary · about