Simulated Annealing

Question 1

What are good tips for finding a global maximum?

Answer 1

• Use a random restart.
• To avoid getting stuck in a flat gradient zone use bigger steps when you find one
• Be careful using big steps because it can make you pass a maximum value or get into an infinite loop.

Question 2

What is annealing?

Answer 2

• Start solving the problem with a lot of randomness and gradually decrease the randomness to converge to an answer.-
• This helps to get out of a local minimum to find a global minimum

Question 3

Local beam search

Answer 3

• Instead of tracking just one particle, we use k particles. At each time frame we look at the randomly generated neighbors of these particles and them and keep the k best ones. This until reaching the goal.
• Share information between iterations.
• In stochastic beam search the succesors are not chosen based only on its fitness, but also some are randomly chosen. This to avoid a local maximum

Question 4

Genetic algoritms

Answer 4

• Is an analogy to natural selection in biology. It uses breeding and mutation to find the optimal answer.
• Uses an evaluation function to estimate the fitness of a solution. Then gives a probability proportional to its fitness for being choose to reproduce
• For some iterations the results of the child one parameter gets changed for some child randomly