Schedule for the CEC2017

Multiobjective Evolutionary Computation has been a major research topic in the field of evolutionary computation for many years. It has been generally accepted that combination of evolutionary algorithms and traditional optimization methods should be a next generation multiobjective optimization solver. Decomposition methods have been well used and studied in traditional multiobjective optimization. It is well known that the Pareto optimal solution set of a continuous multiobjective problem often exhibits some regularity. In this talk, I will describe two multiobjective evolutionary algorithms: MOEA/D and RM-MEDA. Both of them borrow ideas from traditional optimization. MOEA/D decomposes a multiobjective problem into a number of subtasks, and then solves them in a collaborative manner. MOEA/D provides a very natural bridge between multiobjective evolutionary algorithms and traditional decomposition methods. It has been a commonly used evolutionary algorithmic framework in recent years. RM-MEDA makes use of the regularity property to model the distribution of Pareto optimal solutions in the search space, and then generates new solutions from the model thus built. Machine learning techniques can be readily used in RM-MEDA. I will explain the basic ideas behind these two algorithms and some recent developments. I will also outline some possible research issues in multiobjective evolutionary computation.

Currently Evolutionary Computation (EC) is used only occasionally for the design of industrial products for the communications industry. The reasons for this are manifold: Firstly, in most cases domain specific design techniques are available and also much better known to the engineers, because these traditional techniques are taught throughout the universities while EC curricula are rare. Secondly, many EC methods are not simple enough to understand to justify them being applied. Given that convergence of EC is generally not guaranteed, the risk of spending time for an unfamiliar method is not taken. So today EC does not yet belong to the standard “bag of tricks” an engineer is equipped with.Thirdly, undertaking a design by employing EC comes with fairly long convergence times compared to traditional techniques, mostly because the latter are tuned to exploit domain properties and hence are specialized for the problem at hand. In some instances, however, traditional approaches do not yield the desired result or cannot solve the problem at all so that EC techniques have to come to the rescue. I will provide some real-world examples that have been designed with Differential Evolution (DE) and that have entered real products. I will explain why DE has been used in these cases, how the problems were solved, and what is required for EC methods to gain a stronger foothold in industry, hopefully spawning a new research direction to benefit both EC in general as well as industrial applications.

Black box optimization begins with the assumption that if nothing is known about the objective function, then there is no justifiable reason for an optimization algorithm to, e.g. preferentially search in one direction, or to favor one coordinate system orientation over another. This paper investigates whether or not classic differential evolution (DE/rand/1/bin) satisfies these black-box constraints and if not, what it takes to bring the algorithm into conformity with them. The result is an exceptionally simple algorithm, black box differential evolution (BBDE), whose performance is invariant under a coordinate system translation, an orthogonal rotation, a reflection and a permutation of parameters. Performance is also invariant under both the addition of a function bias and an order-preserving transform of the objective function. On the family of ellipsoids, its performance is invariant to both scaling and high-conditioning (eccentricity). Additionally, BBDE is free of both selection and generating drift biases. Furthermore, selection, mutation and recombination are decoupled to become independent operations, as they should be, since each performs a distinctly different function that ought not to be duplicated by another. BBDE also satisfies a few algorithm-specific, symmetry-based constraints. Like the CMA-ES, BBDE’s only control parameter is the population size. In short, BBDE appears to be the simplest DE strategy to conform to a set of symmetry-based constraints that are necessary for unbiased, i.e. black box, coptimization.

Clustering is an unsupervised exploratory data analysis tool which groups the data on the basis of some similarity/dissimilarity metric such that a pre-defined criterion is optimized. The problem of clustering is therefore essentially one of optimization. The use of metaheuristic methods like genetic algorithms has been used successfully in the past for clustering a data set. It is to be noted that the clustering problem admits a number of criteria or cluster validity indices that have to be simultaneously optimized for obtaining improved results. Hence in recent times the problem has been posed in a multiobjective optimization (MOO) framework and popular metaheuristics for multiobjective optimization have been applied. In this talk, we will first briefly discuss about the fuzzy c-means algorithm and the basic principles of MOO. Subsequently it will be shown how a popular multiobjective optimization algorithm may be used for solving the clustering problem. Since such algorithms provide a number of solutions, a way of combining the multiple clustering solutions so obtained into a single one using supervised learning will be explained. The talk will conclude by demonstrating an application of the multiobjective clustering technique on several gene expression data sets. Before explaining the application, a brief introduction to gene expression data will be provided and the importance of clustering gene expression data will be discussed.