混合树估计算法及其应用

Estimation of Mixtures of Trees Algorithm and its Application

提出一种混合树分布估计算法,将一种新的图形模式,混合树图形模式纳入了分布估计算法框架中。算法在进化过程中以选择得到的中间群体为数据集,自动学习混合树模型以描述中间群体中各基因之间复杂依赖关系,并从该模型中抽样得到新一代群体。算法具有更强的自适应性,对进化方向也具有更好的指导作用,模型所具有的聚类特性使得算法能够很好的求解多模态优化问题。将算法实际应用于求解多目标图像识别问题,表现出了良好的优化性能。

This paper presents a novel algorithm named estimation of mixtures of trees algorithm. A new graphical model named mixtures of trees is introduced into the estimation of distribution algorithms paradigm. In the evolutionary process of the algorithm, the inter population selected from the old population is used as the learning dataset for the mixtures of trees learning process. Therefore, complex relationships between genes in the inter population can be represented by the mixtures of trees graphical model and can be utilized to generate new population by sampling. The algorithm is self-adaptive, and can guide the search with the model more efficiently. Furthermore, the clustering property encode by the hidden variable facilitate the algorithm to solve multi-modal optimization problems. A practical application of the algorithm to the multi-objective image recognition is introduced, and gives a good demonstration of the performance of our algorithm.