An Evolutionary Approach to Multiobjective Clustering IEEE TRANSACTIONS
An Evolutionary Approach to Multiobjective Clustering IEEE TRANSACTIONS ON EVOLUTIONARY COMPUTATION, VOL. 11, NO. 1, 2007 Julia Handl and Joshua Knowles Speaker: 陳進和 2007年 8月1日
Outline n n Introduction MOCK (Multiobjective clustering with automatic k-determination) Experimantal results Conclusion
1. Introduction n Assess the performance of clustering algorithm n n n Lack of a formal definition of clustering No objective performance criterion Multiobjective optimization is used to tackle n n Unsupervised learning problem Data clustering
2. MOCK (Multiobjective clustering with automatic k -determination) n MOCK consists of two phases n n Phase 1: Initial clustering phase Phase 2: Model-selection phase
Phase 1: Initial clustering phase n PESA-II n n n Internal population to explore new solutions External population to exploit good solutions Objective functions
Phase 1: Initial clustering phase n (cont. ) Genetic representation and operators n n n Locus-based adjacency representation n No need to fix the number of clusters n Well-suited for standard crossover operators Uniform crossover n One-point or two point Neighborhood-biased mutation operator n Quickly discard unfavorable links n Explore feasible solutions
Phase 2: Model selection n Motivating concepts n Inspired by Tibshirani et al. ’s Gap statistic, a statistical method to determine the number of clusters in a data set
3. Experimantal results n Parameter setting
4. Conclusion n MOCK n n Outperform traditional single-objective clustering techniques Keeping the number of clusters dynamically
- Slides: 15