Departmental Colloquium
Jun Liu
Harvard
On the detection of non-independence
Abstract: I will discuss a few recent results from my group aiming to the detection
of non-linear dependence between two random variables. Our approach is
based on an optimal slicing (discretization) of one or both variables to
optimize a score function derived from a likelihood-ratio test formulation.
Our approaches are compared with some well-known methods such as Distance
Correlation, Pearson Correlation, Maximal Information Criterion, etc., on
many simulated examples, and found superior for highly nonlinear and
non-smooth relationships between the two variables. We will also show how
these methods are applied to bioinformatics problems such as gene-set
enrichment analysis, transcription regulation analysis, etc.
Friday November 14, 2014 at 3:00 PM in SEO 636