### Aaditya Ramdas - Interactive multiple testing with STAR: selectively traversed accumulation rules for structured FDR control

205
views
0
7 months ago by
Community: WHOA-PSI 2017

0
7 months ago by
Updated slides.

File attached: STAR-PDF.pdf (2.59 MB)

0
7 months ago by
I guess I had in mind a (related?) polynomial regression model with a heredity property of the truth, in that if a polynomial in covariate $X_{j}$ of degree $k$ is included in the model, then all polynomials of degree $d<k$ are also included. Specifically, say $X_{4}^{3}$ is a true active predictor, then $X_{4}^{2}$ and $X_{4}$ are also in the true active set.
0
7 months ago by
If you can get independent p-values for these, our STAR method is applicable. Otherwise, I have a different paper coming out soon on the first known (non-interactive) procedure for FDR control for DAGs, that can handle positive or arbitrary dependence between the p-values.
0
7 months ago by
Thank you, Aaditya, for the great presentation.  If you are presenting this project to a more applied audience, I would recommend describing it as an automatic technique from the beginning (as you mentioned in the end).  Good luck on the job market!