摘要
针对传统聚类模型的缺陷，文章利用确定性退火技术，提出一种聚类模型及聚类算法．该模型考虑了聚类的交互作用，以前提出的一些聚类模型是它的特例．引入温度参数，把聚类问题看成一个物理系统，把求解聚类问题的最优解转化为模拟随温度变化的物理系统的平衡态．通过求解一系列随温度变化的物理系统的自由能函数的局部极小来模拟物理系统的平衡态，最终达到物理系统的基态，即聚类问题的最优解．
Aiming at the defects of traditional clustering model, a kind of clustering model and algorithm are put forward and researched by use of deterministic annealing. The model takes account of the interactions of clusters, some models which were put forward previously are special cases of this one. Temperature parameter is introduced, and the clustering problem as a physical system is considered. Finding the optimal solution to clustering problem is transformed into simulating the equilibrium state of a physical system. The equilibrium state is simulated by solving a series of problems to minimize the free energy which varies with temperature, and finally, the ground state of the system is attained. That is the optimal solution of clustering problem.
出处
《软件学报》
EI
CSCD
北大核心
1999年第6期663-667,共5页
Journal of Software
关键词
确定性退火
聚类
算法
人工智能
模式识别
Deterministic annealing, clustering, free energy, the principle of maximum entropy.
