一种自适应阻尼因子的仿射传播聚类算法

Affinity propagation clustering algorithm with adaptive damping factor

仿射传播聚类算法已经被广泛应用于各个领域,其源码被Toronto大学公开在网络中。针对公开源码中如何选择阻尼因子的值以平衡算法震荡与收敛速度的问题,提出一种自适应阻尼因子的仿射传播聚类算法。所提算法通过监视算法的震荡情况,自适应调整阻尼因子的值,相比公开源码中的固定阻尼策略,不仅可以有效避免震荡,且可以很大程度地保持阻尼因子较小时的收敛速度。通过多个UCI公开数据集试验证明了所提算法的有效性。

Affinity propagation clustering algorithm has been extensively applied in various fields. Its source code is opened on the network by Toronto University. Aiming at how to choose the damping factor value of open source to balance the convergence speed and oscillation problems of the algorithm, an affinity propagation clustering algorithm with adaptive damping factor is proposed. The proposed algorithm adaptive factor adjusts the damping factor value by monitoring oscillation situation of the algorithm. It can not only effectively avoid the oscillation, but also greatly keep the convergence speed with the small damping factor compared with the fixed damping factor strategy in open source. Experiments using multiple UCI public data show the validity of the proposed algorithm.