MISTA 2011: Programme
The list of accepted papers and program schedule can be downloaded from here.
We are fortunate enough to have the following plenary speakers at the conference
, IBM Systems and Technology Group
The Ongoing Challenge: A Responsive Enterprise-Wide Demand Supply Network for Semiconductor and Package Operations - Something for Everyone
Abstract: Most organizations, from health care facilities to semiconductor manufacturing, can be viewed as an ongoing sequence of loosely coupled decisions where current and future assets are matched with current and future demand across the demand-supply network at different levels of granularity ranging from placing a lot on a tool to an aggregate capacity plan across a five-year horizon. By itself, this creates a substantial challenge to management and the computational intelligence community to in place applications that enable a firm to respond quickly and intelligently to changes in demand and/or assets. The nature of semiconductor manufacturing adds such features as re-entrant flow, alternative bill of material, mixing process and assembly operations, long and short lead times, variability in demand, etc to the challenge pile
. This presentation will outline these challenges at the global level, the factory level, and coordination between these two levels covering the basic playing field, the evolution of approaches, some current best practices, and emerging challenges - something for everyone.
, Lancaster University, UK (also see here
Stochastic scheduling: what everyone should know about index policies
Abstract: A multi-armed bandit problem concerns N >= 2 independent populations of rewards whose statistical properties are unknown (or at least only partly known). A decision-maker secures rewards by sampling sequentially from the populations, using past sampled values to make inferences about the populations and so guide the choice of which population to sample next. The goal is to make these choices in such a way as to maximise some measure of total reward secured. Such problems embody in a particularly simple form the dichotomy present in many decision problems between making decisions with a view to securing information which can improve future decision-making (exploration) and those which exploit the information already available (exploitation). In the 1970's John Gittins discovered that important classes of such multi-armed bandit problems have solutions of a particularly simple form: at each stage of the sampling compute an index (the Gittins index) for each of the N populations, namely a function of the rewards already sampled from the population concerned. Always sample next from the population with the largest index. Moreover, the index concerned has a simple interpretation as an equivalent known reward for the population concerned. It emerges that many problems involving the sequential allocation of effort, some of quite different character to the above multi-armed bandit problems, have index solutions. Since the 1970's, Gittins' index result together with a range of developments and reformulations of it have constituted an influential stream of ideas and results contributing to research into the scheduling of stochastic objects. Application areas to which these ideas have contributed include approximate dynamic programming, the control of queuing systems, fast fashion retail, machine maintenance, military logistics, optimal search, research planning, sensor management, communication channel usage and website morphing. The talk will give an overview of some key ideas and some recent developments.
, Colorado State University, USA
Flying High, or Under the Radar? Applications of Evolutionary Algorithms in Optimization, Search and Scheduling
Abstract: Evolutionary Algorithms have been part of main stream research in Computer Science and Operations Research for 20 years. Yet, there does not seem to be many highly visible applications of Evolutionary Algorithms. For example, it is not widely known that the design of the GE engines currently flying on the Boeing 777 were evolved using a genetic algorithm. This talk will explore significant yet relatively unknown applications of evolutionary algorithms, and will explore the question as to why many applications of evolutionary algorithms are not well known. The talk will look specifically at scheduling applications and will try to sort out subdomains were evolutionary algorithms work well and subdomains where they do not work well. The role of benchmark test problems and how they differ from real world scheduling applications will also be considered.
We will publish the rest of the program once it is available.