Department of Computer, Control and Management Engineering

Giovedi' 3 dicembre ore 12 Aula Magna:
Sean Luke, Department of Computer Science, George Mason University
Metaheuristics, Evolutionary Algorithms, and Coevolution

Abstract
Metaheuristics is the common but unfortunate name for stochastic
optimization algorithms such as Simulated Annealing, Ant Colony
Optimization, and the Genetic Algorithm.  These are highly general,
model-free methods of last resort before you are forced to use brute-force
or random search.  Metaheuristics are often the only known way to solve a
great many problems, but typically incur a very high computational cost.
Probably the most well-known metaheuristics are sample-based methods
collectively known as Evolutionary Algorithms (such as the Genetic
Algorithm), which are popular because they are very readily parallelizable
and adaptable to multiobjective optimization and large problems.

Sean Luke will introduce the general concept of metaheuristics, with a focus
on evolutionary algorithms and their application, and will then discuss his
recent work in evolutionary algorithms, with a focus on coevolution.
Coevolution is a multiagent form of evolutionary algorithms which
significantly reduces the complexity of large
problems by enabling a form of partial problem decomposition.
Game-theoretic results in coevolution have had significant recent impact on
other areas, including multiagent reinforcement learning and estimation of
distribution algorithms.