Simulated Annealing
Search that accepts worse answers on purpose, less and less often, to avoid getting stuck.
Why It Was Important
Borrowed from metallurgy, where slow cooling lets atoms settle into a low-energy arrangement, simulated annealing lets an optimiser move downhill sometimes — often at first, then rarely — so it can escape a local optimum instead of settling in the first one it finds. It gave optimisation a general-purpose escape mechanism, and its statistical-physics framing carried directly into the Boltzmann machine, which Hinton and Sejnowski built on the same energy-landscape idea two years later.
Who Invented It
Scott Kirkpatrick, C. Daniel Gelatt, Mario P. Vecchi
IBM researchers who adapted the Metropolis algorithm from statistical physics to combinatorial optimisation.
Applications
- Combinatorial Optimisation
- Circuit Design
- Scheduling
- Neural Network Training
Key Papers
- Optimization by Simulated Annealing
Scott Kirkpatrick, C. Daniel Gelatt Jr., Mario P. Vecchi · Science · 1983