Ensemble Based Simulated Annealing

July 24, 1994

Sheraton Harbor Island East

San Diego, California

Tutorial Description

The past ten years of research has led to adaptively optimized implementations of simulated annealing. The theory and techniques for such fine tuning (adaptive cooling schedule, ensemble size, and stopping temperature) can result in an order of magnitude improvement in the performance of simulated annealing algorithms.

The course will introduce these techniques and offer hands-on experience with the new software package EBSA. It will include presentations on the theory of optimally adaptive simulated annealing followed by hands-on training. SDSC EBSA has been written to run in a generic Unix environment and optimized for all platforms at SDSC: Cray YMP and C90; Intel Paragon; plus DEC SGI, and Sun workstations. The visualization software implemented in EBSA is currently only compatible with SGI GL software laboratories. An X-window release is scheduled for later this year.

The presentations will address issues concerning the optimal implementation and provide comparison with other approaches to global optimization including genetic algorithms, evolutionary programming, and neural networks. The approach described in the course and embodied in the software represents the state-of-the-art technique of global optimization as currently practiced in many academic and industrial groups around the globe.

A color video projector will be used for live demonstrations during the training sessions. A network of graphic workstations will also be provided for hands-on training. Sample problems, programs, and software documentation will be provided. Participants are encouraged to bring problems of interest to them on SunSparc 3.5" disks or SGI 4mm DAT tape.

The course is 60% introductory, 35% Intermediate, and 5% advanced.

Who Should Attend

Graduate students, scientists, engineers and mathematicians interested in using simulated annealing, in global optimization or application of simulated annealing to practical problems.

Recommended Background

Graduate standing in computer science, mathematics, or physics.

Instructors:

Richard Frost, San Diego Supercomputer Center and Peter Salamon, San Diego State University

Richard Frost is a Senior Analyst at the San Diego Supercomputer Center where he works on computational problems in applied mathematics and parallel computing. His experience includes software implementation on SIMD, VP, and MIMD supercomputers along with standard workstations and PC's. He is currently chair of the Intel Supercomputer Users Group Visualization SIG. Recent publications include work in materials science, stochastic optimization, and signal processing.

Peter Salamon has been full professor of applied mathematics since 1984 in the Department of Mathematical Sciences, San Diego State University. He has taught graduate and undergraduate courses on optimization theory at SDSU and courses on simulated annealing at the University of Copenhagen and at the Danish Technical University. He has numerous publications on simulated annealing. Currently he is on a year's sabbatical at the Computational Neural Network Center in Denmark.


Program

Sunday, July 24, 1994

Theory/Harbor 3

Hands-On Sessions

Note: The instructors will answer individual questions during all hands-on sessions. There will be ample time for discussion. The tutorial will take place in Harbor 3. Coffee and lunch will be in Harbor 2.