Tel-Aviv University - Computer Science Colloquium
Sunday, May 30, 14:15-15:15
at 14:00 Modeling autocorrelated stochastic processes is difficult. Examples include security prices and bursty teletraffic in emerging telecommunications network, especially video traffic. Performance evaluation of the latter case calls for queueing systems with bursty (autocorrelated) arrival traffic, typically analyzed by simulation. The importance of queues with autocorrelated arrivals and services is that they give rise to performance measures which are very different, and typically much worse, than those predicted by classical queueing with renewal (GI) arrivals and service. In particular, positively autocorrelated interarrival intervals give rise to bursty arrival traffic which can dramatically degrade customer-oriented performance measures, such as waiting times and loss probablities.
We present a class of methods called TES (Transform-Expand-Sample) for generating autocorrelated random sequences. TES processes use autoregressive modulo-1 schemes, with additional memoryless transformations. They are readily implemented on a computer, have a low computational complexity and give rise to a variety of autocorrelation functions, including monotone as well as oscillatory ones. Since their autocorrelation functions can be computed by fast numerical algorithms (instead of simulation-based statistical estimates), TES modeling can be accomplished via heuristic search over a suitable parameter space. An algorithmic search procedure has also been developed. The Empirical TES Modeling methodology is a novel input analysis paradigm, primarily for Monte Carlo simulations, that constructs a TES model from empirical time series data (measurements); the constructed TES model is a stationary time series which matches the empirical marginal distribution (histogram) and simultaneously approximates the empirical autocorrelation function.
This talk will outline the Empirical TES Modeling methodology and present case studies from a broad range of application domains. We demonstrate how the methodology was used to construct source models of VBR (variable bit rate) compressed video for QoS (quality of service) assessment, fault arrivals for reliability and quality control, financial time series for generating financial scenarios and estimates, and graphical textures. It will conclude with a brief description of a TES-based modeling software package, called TEStool, designed to support visual interactive heuristic TES model searches, as well as automated TES modeling. If possible, a hands-on demo will accompany the talk.
Bio:
Benjamin Melamed is a Professor and Vice-Chair at the Rutgers University Faculty of Management, Department of MSIS, and a member of RUTCOR. His research interests include system modeling and analysis (especially telecommunications systems and traffic), simulation, stochastic processes and visual modeling environments. He authored or co-authored over 80 papers and a book. Melamed was awarded an AT&T Fellow in 1988. He was elected IEEE Fellow in 1994 and an IFIP WG7.3 member in 1997. He was elected to Beta Gamma Sigma in 1988.
Melamed received a B.Sc. degree in Mathematics and Statistics
from Tel Aviv University in 1972, and a M.S. and Ph.D. degrees
in Computer Science from the University of Michigan in 1973
and 1976, respectively.
From 1977 to 1981 he taught at the department of Industrial
Engineering and Management Science at Northwestern University.
He joined the Performance Analysis Department at Bell Laboratories
in 1981, and later became an AT&T Fellow.
Melamed moved to NEC in 1989 and served there as a Deputy Director,
Head of the Performance Analysis Department, and NEC Fellow.
He consulted at Bellcore in 1995 and joined Rutgers University
in 1996.
For colloquium schedule, see http://www.math.tau.ac.il/~matias/colloq.html