Professional Documents
Culture Documents
Simulated Annealing
Rohit Man Amatya
Master by Research Kathmandu University Dhulikhel, Kavrepalanchok
2012, July 20
Simulated Annealing
1 of 21
Introduction Annealing Simulated Annealing Examples Travelling Salesman Problem Ball on Terrian Analysis Requirement Update Function
Simulated Annealing
2 of 21
Problem
We need a strategy to nd other minima This means, we must sometimes select new point How?
Simulated Annealing
3 of 21
Methods
Simulated Annealing
4 of 21
Metal Annealing
Heating to high temperature Gradual cooling Done for changing the properties of solid-materials.
Simulated Annealing
10 of 21
What happens
Change the property of the metal One manner is which crystal are formed low energy, organized crystal lattice formed. Placing themselves in a pattern which is global energy minimum of a perfect crystal.
Simulated Annealing
11 of 21
Simulated Annealing
Incorporate a temperature parameters into the minimization procedure At high temperatures, explore parameter space At low temperature, restrict exploration
Simulated Annealing
12 of 21
Strategy
Consider decreasing series of temperatures For each temperature, iterate these steps:
Propose an update and evaluate function Accept update that improve solution Accept some updates that dont improve solution
Acceptance probability depands on temperature parameter
Simulated Annealing
13 of 21
Salesman must visit every city in a set Given distance between pairs of cities Find the shortest route through the set
Simulated Annealing
14 of 21
Intially placed at a random can only move one step to settle at the lowest point on the terrain
Simulated Annealing
15 of 21
Irreducibility
Simulated Annealing
16 of 21
Accepting an Update
min 1, e T
Simulated Annealing
17 of 21
Equilibrium Distribution
P(S = k ) e T
At low T , probability is concentrated in low energy states
Simulated Annealing
18 of 21
Update Scheme
Simulated Annealing
19 of 21
Applications
Circuit partitioning and placement. Strategy scheduling with complex product structure. Event-based learning situations.
Simulated Annealing
20 of 21
Issues
Selecting the parameters Hard fast rule doesnt exist. Large no of iteration need to gurantee convergence.
Simulated Annealing
21 of 21