[an error occurred while processing this directive]

Efficient Simulation for Rare Events and Combinatorial Optimization

Center for Telematics and Information Technology (CTIT)
University of Twente
July 15-16, 1999

For location and time schedule, click here.

LECTURER

Professor Reuven Y. Rubinstein

William Davidson Faculty of Industrial Engineering and Management
Technion, Haifa, Israel
Telephone: 972-4-8294458
Fax: 972-4-823-5194
E-mail: ierrr01@ie.technion.ac.il
WWW: http://iew3.technion.ac.il:8080/ierrr01.phtml

SCOPE

Stochastic systems abound in real-world applications. Examples include traffic systems, flexible manufacturing systems, computer-communications systems, inventory systems, production lines, coherent lifetime systems, PERT networks and flow networks. Most of these systems can be modeled in terms of discrete events whose occurrence causes the system to change from one state to another. Such systems are called `discrete-event systems' (DES). For most real-world DES, analytical methods are not available and they must be studied via simulation. In designing, analyzing and operating such complex DES, one is interested, however, not only in performance evaluation but also in sensitivity analysis and optimization. Sensitivity analysis is concerned with evaluating sensitivities (gradients, Hessians, etc.) of performance measures with respect to parameters of interest. It provides guidance for design and operational decisions and plays a pivotal role in identifying the most significant system parameters, as well as bottleneck subsystems. Optimization is concerned with a DES as a whole; in particular, it makes use of sensitivity analysis to find the optimal solution with respect to parameters of interest.

The goal of this course is to present a number of efficient (smart) Monte Carlo techniques for performance evaluation, sensitivity analysis with an emphasis to probability of rare event estimation in stochastic networks. In particular the audience will learn:

  1. How a single simulation run can be used to estimate an entire response surface and the associated gradient (sensitivity) curves in an efficient way.
  2. How the calculation of simulation output statistics can be speeded up by factors of tens and hundreds, for typical performance measures, and by factor of thousands, for rare event simulations.
  3. How to solve combinatorial optimization problem, like the maximal cut problem, traveling salesman problem, etc., using efficient rare event techniques and information theory.
It will be shown that all the above are feasible due to the fact that the simulation experiments to be performed on a digital computer are largely computational and not purely statistical. It will also be explained how to use the cross-entropy method to extract the most valuable information obtained during the course of simulation.

A research software package demonstrating the efficacy and efficiency of the proposed techniques will be made available for the participants of this course, as well as a number of case studies. Participants are encouraged to bring along their own examples and case studies and run them during and after the course.

WHO SHOULD ATTEND

Simulation practitioners, computer scientists, industrial, electrical and financial engineers, graduate students and anyone interested in fast simulation of discrete event systems.

TOPICS OUTLINE

  1. Fast estimation of performance measures for both static and dynamic models via simulation.
  2. Fast estimation of derivatives (sensitivities) from a single simulation run.
  3. Fast estimation of probabilities of rare events via exponential change-of-measure and cross-entropy.
  4. Combinatorial optimization problems via rare events and cross-entropy.
  5. Case studies and software presentation.

EXPECTED BENEFITS

Participants will learn the key ideas and features of modern simulation. Special emphasis will be placed on sensitivity analysis, probability of rare-event estimation and randomized algorithms for combinatorial optimization. Participants will be shown how to take advantage of simple change-of-measure ideas and cross-entropy so as to reduce the time complexity required by statistical computations in Monte Carlo simulations. These ideas will then extended to the speeding up of computations in stochastic and deterministic combinatorial optimization models.

PARTICIPATION

All interested are invited to participate at no charge! Each participants is responsible for his own arrangements to attend the course (e.g., accommodation, meals, etc.) For admission (on a FCFS basis) and other details, please e-mail to: resim99@cs.utwente.nl.


PROGRAM AND TIME SCHEDULE

Thursday, July 15th, 10:00-12:30 and 14:00-17:30 :
Friday, July 16th, 9:00-13:00 :

LOCATION

Room INF-U2,
Computer Science building ("Informatica-gebouw")
University of Twente, Enschede, The Netherlands

Travelling information to the university is available here.
A map of the campus showing the "INF-gebouw" is available here.


[an error occurred while processing this directive]