Adaptive direction choice in random search continuous global optimization
Danny M. Kaufman, Robert L. Smith · OSTI OAI (U.S. Department of Energy Office of Scientific and Technical Information) · 1994
The Hide-and-Seek continuous simulated annealing algorithm proceeds iteratively by a random step size in a randomly chosen direction. The probability distribution of direction choice has been shown to affect the rate of convergence to the stationary Boltzmann distribution corresponding to a fixed temperature. We review direction choice rules involving estimation of Hessian information, motivated by attempts to optimize known bounds on the rate of convergence. Prior work has used Hessian information to estimate the local shape of the objective function. We can also use gradient information to bias the direction distribution according to our current position within the local basin.