A Kernel Statistical Test of Independence

Although kernel measures of independence have been widely applied in machine learning (notably in kernel ICA), there is as yet no method to determine whether they have detected statistically significant dependence. We provide a novel test of the independence hypothesis for one particular kernel independence measure, the Hilbert-Schmidt independence criterion (HSIC). The resulting test costs O(m2), where m is the sample size. We demonstrate that this test outperforms established contingency table and functional correlation-based tests, and that this advantage is greater for multivariate data. Finally, we show the HSIC test also applies to text (and to structured data more generally), for which no other independence test presently exists.

Authors

• Arthur Gretton University College London
• Kenji Fukumizu
• Choon Teo
• Le Song
• Bernhard Schölkopf
• Alex Smola

Keywords

No keywords are indexed for this paper.