Information-Based Global Optimization

• Choose initial point and evaluate

• Iterate:

  • Pick next point according to g and evaluate

• Problems: convergence, points clustered near minima

Previous slide

Next slide

Back to first slide

View graphic version

The simplest form of information-based global optimization is then described as above.

This graph demonstrates information-based global optimization of a parabola minimized at the origin. The randomly selected initial point is in the lower left. Given the lower left value, a zero is sought in the upper right, the upper left, the lower right, and then close to the center. Then we see the disadvantage of sampling in the decision procedure. The subsequent 25 points do not converge to the origin and the decision procedure is bogged down by a cluster of points which don't provide much global information to the global decision procedure. It would be desirable to summarize such information and speed convergence.