Search Without a Gradient

Find the bottom of a bowl you cannot see: measure the slope where you stand, step against it, repeat. The learning rate decides whether you crawl, arrive, bounce, or fly off.

A walker on a bumpy landscape accepts uphill moves with a chance that shrinks as the temperature cools, so it escapes small basins early and settles late. At zero temperature it is hill climbing.

A climber with a memory: take the best neighbour even when it is worse, forbid undoing the move for a few rounds, and break the rule only for a new best. With no list it bounces between two states.

A population of bit strings packs a knapsack: tournaments pick parents, a one-point cut splices them, rare flips mutate, the best survives untouched. Compare it with the exact knapsack table.

Ants build tours city by city, each next step a gamble weighted by pheromone and closeness; trails evaporate and the shortest tours lay the most, so the colony converges on good routes.

Whales circle a leader, spiral in, or wander, with a coefficient that shrinks from 2 to 0. A clear look at a 2016 metaheuristic whose moves, critics showed, are old operators under a new story.

A class learns from its best student and from each other: a teacher phase pulls everyone up, a learner phase pairs them off, and only better moves are kept. Billed parameter-free; read the critique.