NIPS2009_0215_slide - 18000 20000 22000 1150 1200 1250 1300...

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
Decoupling Sparsity and Smoothness in the Discrete HDP Chong Wang and David M. Blei Computer Science Department, Princeton University Problem : I In hierarchical Dirichlet process (HDP), the topic sparsity and smoothness is coupled through a single parameter, making it hard to achieve both. Solution : I Sparse topic models (sparseTMs) decouple the sparsity and smoothness using selector variables to choose the terms for topics. I This finds simpler models which achieve better predictive perplexities.
Background image of page 1
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: 18000 20000 22000 1150 1200 1250 1300 arXiv complexity perplexity stm hdp-lda ● ● ● ● ● ● ● ● ● ● ● ● ● 16000 17000 18000 19000 700 750 800 850 Nematode Biology complexity stm hdp-lda ● ● ● ● ● ● ● ● ● ● ● ● ● ● 34000 40000 46000 1350 1450 NIPS complexity stm hdp-lda ● ● ● ●● ● ● ● ● ● ● ● ● 19000 21000 23000 920 960 1000 1040 Conf. abstracts complexity stm hdp-lda ● ●...
View Full Document

This note was uploaded on 02/12/2010 for the course COMPUTER S 10586 taught by Professor Jilinwang during the Fall '09 term at Zhejiang University.

Ask a homework question - tutors are online