|
|
|||||||||||||||||||||
|
Topology seminar
Speaker: Lekha Das, IIT Bombay
Host: Rekha Santhanam
Title: Admissible sets and N_infinity operads
Time, day and date: 2:30:00 PM – 3:30:00 PM, Thursday, July 30
Venue: Room 215
Abstract: We discuss the admissible sets for linear isometries operad and little disk operad, and we explain the relationship between them.
Statistics and Probability seminar
Speaker: Sagnik Nandy, Ohio State University
Host: Parthanil Roy
Title: How does limma-trend work? An empirical partially Bayes perspective
Time, day and date: 4:00:00 PM - 5:00:00 PM, Thursday, July 30
Venue: Ramanujan Hall
Abstract: In high-throughput biology, we often want to test thousands of genes or proteins at once, but each one is measured with very few samples — so noise estimates are unreliable, and standard tests find almost nothing. A widely used fix, called limma-trend, works by "borrowing strength" across all the measured units: instead of trusting each unit's own noisy variance estimate, it shrinks that estimate toward a shared trend learned from the whole dataset. This talk explains what limma-trend is actually doing statistically, why it works, and when it can fail. We then introduce two more flexible versions of the method that adapt automatically to the data rather than assuming a fixed mathematical form and prove that they reliably control the false discovery rate — even when the underlying trend is estimated imperfectly. On real cancer genomics and proteomics datasets, these new methods substantially increase the number of genuine findings recovered, while avoiding a bias problem that affects several existing tools. (This talk will be based on a joint work with Wanyi Ling and Nikos Ignatiadis.)