TAM-EDA: Multivariate t Distribution, Archive and Mutation Based Estimation of Distribution Algorithm

Abstract

We present a novel estimation of a distribution algorithm (eda), tam-eda, which uses a multivariate t distribution model, an archive population and a mutation operation to escape local minima, avoid premature convergence and utilize a record of the best solutions. Earlier edas used multivariate normal distributions to model low-cost regions of the search space. The multivariate t distribution has heavier tails and so is more likely to maintain diversity, while still allowing convergence to occur. The current population of potential solutions has limited ability to represent all the best regions of the search space explored so far. The archive allows storage of a larger population of promising solutions, which are then used in model building. However, the eda Contents C721model and archive may still become stuck at suboptimal solutions, so to combat this we introduce a decomposition mutation operation which retains most of the attributes of a current solution but attempts large changes in others. A comparison with generic eda, genetic algorithms and the Nelder-Mead method shows that tam-eda is an effective optimization algorithm for a range of test problems.

Large-Scale Estimation of Distribution Algorithms with Adaptive Heavy Tailed Random Projection Ensembles

“…The increased exploration abilities of heavy tailed distributions are well known in evolutionary search, starting from pioneering work by Yao et al [19] in the univariate setting, and more recent work in the multivariate EDA framework -see e.g. [20][21][22] and references therein. However, in the regime of large search spaces of dimensionality well beyond 100 variables, the use of such heavy tailed distributions in EDA is not straightforward -as demonstrated in [22], a heavy tailed search distribution becomes increasingly counter-productive as it loses sight of the direction of the search.…”