Presentation of Evolution and Algorithms
summary
Shows that standard population genetics under weak selection implements the multiplicative weights update: allele frequencies follow the MW equations, so expected fitness improves while trading off entropy, and regret to the best “expert” is about 1/sqrt(T). Gives specific scaling: Kimura-style information accrues ~C per generation and adaptations plausibly occur every ~300–400 generations. Proves a complexity claim for random pairwise-interaction fitness landscapes: PLS-hard: exponential convergence (even infinite populations need exponential time or size to fix local optima).
AI-written summary · about