Simulated Annealing
How does it work?
Simulated Annealing methods search a space of candidate solutions. They typically define neighbor moves or gradients, evaluate objective functions, and use schedules or memory to escape local optima or to converge reliably.
Examples
- VLSI placement — Optimise chip component layouts with a cooling schedule to escape local minima.
- Traveling Salesman approximations — Find near-optimal tours via random neighbour moves and temperature-controlled acceptance.
- Job-shop scheduling — Schedule tasks on machines by accepting worse moves early and reducing acceptance over time.
Problems
- Designing an effective cooling schedule
- Slow convergence relative to other metaheuristics
- Sensitive, problem-specific tuning of the acceptance function
- No guarantee of finding the global optimum in finite time
- Hard to parallelize due to its sequential nature