maximum margin dirichlet process mixtures for clustering, refer to https://www.aaai.org/ocs/index.php/AAAI/AAAI16/paper/download/11828/11763
it is a discriminative model for nonparameteric clustering, which leverages dirchlet process and maximum margin clustering
#Gibbs sampling:
- it uses dirichlet process as the prior to generate the number of clusters
- the likelihood is from maximum margin model
- inference is done based on posterior probability
it is based on maximum margin online learning to update component parameters
Check the parameter C in dpmm_mmc.m, which balances the contribution from prior and likelihood.
#Demo demodpmm_mmc.m
#Reference Maximum Margin Dirichlet Process Mixtures for Clustering, G. Chen, H. Zhang and C. Xiong. AAAI Conference on Artificial Intelligence (AAAI 2016).