At the forefront of Artificial Intelligence

Home » Articles » Neural Networks » Applications/Code

Simulated Annealing Demonstrator (SAD)

Simulated Annealing is a method of avoiding local minima. Annealing is a process that involves heating metal to incredibly high temperatures and slowly cooling it down so that the atoms form strong bonds between them. How does this help Artificial Intelligence? When applied to neural networks, it can help bump them out of local minima in their energy surfaces.

This example does not use neural networks, but simply tries to minimize the following function:

Notice how the bottom of the function includes two little bumps. Using mathematical functions, you can determine the minimum to be around (0.3151074078, -8.382084455). When you run the program, if you run it multiple times you will see the algorithm jump back and forth between the two humps. More temperatures and a greater number of iterations would probably reduce the possibility of choosing the wrong minima.

See the Generation5 essay for more details.

Submitted: 18/09/2000

Article Toolbar

BibTeX entry