search
Climb out of the local trap
It is the era when some optimization landscapes are simply too rugged for any method that only ever moves downhill: hill-climbing marches straight into the nearest local optimum and stops, blind to the far better solutions across the next ridge. Design a search that escapes — one that, when stuck, does not surrender but perturbs, restarts, or accepts a lateral or uphill step to break free, trading the guarantee of monotone improvement for the ability to roam. For hard satisfiability and layout problems this stochastic local search often crushes exhaustive methods in wall-clock time, without any guarantee it will. State honestly what you give up. Get it wrong and you freeze at the first local optimum, or wander forever — balancing exploration and exploitation is the whole game.
Who this problem belongs to
The two figures whose methods fit it best, out of 70 in contention.
Moore's research on phase transitions in satisfiability and community detection is the direct scientific study of why some optimization landscapes are riddled with local optima that trap simple hill-climbing, and why, near the hard region, escaping them requires exactly the kind of perturb-restart-or-accept-a-worse-step strategy this problem describes. His Santa Fe Institute physics-of-computation lens explains why stochastic local search sometimes crushes exhaustive methods in wall-clock time without offering any guarantee, precisely the honest tradeoff this problem demands students state. Few researchers on this roster have written more carefully about the statistical-physics view of rugged landscapes (energy surfaces with many local minima) that underlies simulated annealing and WalkSAT-style methods. His score is not higher only because he is a theorist of the landscape rather than the inventor of a specific escape algorithm.
Hopfield's energy-based neural networks are literally a physical model of a rugged optimization landscape with many local minima (attractor states), and his work connecting statistical physics to computation is the direct intellectual ancestor of simulated annealing, which borrows explicitly from spin-glass physics to explain why accepting occasional uphill moves helps a search escape a shallow local trap. His framework gave later researchers (Kirkpatrick, Gelatt, Vecchi in 1983) the physical metaphor for exactly the strategy this problem asks for: perturb, sometimes accept worse, and let a temperature-like parameter control the tradeoff between exploration and exploitation. He did not himself formalize simulated annealing or stochastic local search as general algorithms, so his score reflects the deep physical intuition underlying the method rather than direct authorship of the escape strategy itself.
In the mind map
The same ideas, as concepts rather than history — in John's ML knowledge map.
70 figures are scored on this problem. Draw it in a battle to see where you land.