Kernel Measures of Independence for non-iid Data
Xinhua Zhang (ANU & NICTA), Le Song (CMU) Arthur Gretton (MPI Tübingen), Alex Smola (Yahoo! Research)


W 39

Problem: Extend HSIC to structured non-iid Data
Approach: Exponential Family Models
 Injective  Decompose on graphical models



Marginal Polytope in an RKHS


Estimator: unbiased, deviation is asymptotically normal Assume: homogeneity, stationarity





Applications: Independence test ICA (contrast function) Time series segmentation