COMP9844 Extended Neural Networks

Simulated Annealing - Solutions

It is best not to read the answers until you've tried to answer the questions yourself.

  1. Describe the concept of an annealing schedule.

    Answer: An annealing schedule is a scheme for lowering the simulated temperature from time to time, after allowing the system to reach equilibrium at each new temperature.

  2. Describe the essence of the Metropolis algorithm.

    Answer: In the Metropolis algorithm, a unit is chosen at random and subjected to a small displacement, and the resulting change in energy ΔE of the system is computed. If ΔE ≤ 0, the displacement is unconditionally accepted, while if ΔE > 0, the displacement is accepted with probability pexp(–ΔE/T, where p is a constant and T is the pseudo-temperature. This process is repeated until the system reaches equilibrium.

  3. What is the purpose of allowing displacements that increase the energy E in an algorithm that aims to minimise E?

    Answer: This is to allow the system to escape from local minima of the energy landscape.


Copyright © Bill Wilson, 2009.
Bill Wilson's contact info

UNSW's CRICOS Provider No. is 00098G