Please join us for the Statistics Seminar scheduled from 11 a.m. to noon on Friday, April 5.
Our distinguished guest speaker, Ashis SenGupta, CSIR Emeritus Scientist, Govt of India, will be present a talk on "Probability Distributions on Manifolds with Applications to Statistical Machine Learning & Bioinformatics."
The seminar will be conducted in-person in HOS 380; however, virtual attendance is also available via Zoom. Further details can be found below.
- Zoom link
- Meeting ID: 935 6604 6886
- Passcode: 25396015
Abstract:
Non-linear data are fast emerging in a variety of modern applications of applied sciences. Applications of the usual methods for analyses of linear data on $\mathrm{R^1}$ to those for such non-linear data as with circadian rhythms, protein angles, chrono-therapy, etc. are flawed. Such applications lead to misleading inferences with possibly devastating consequences, especially when dealing with medical problems in human beings. As a simple example with data from an actometer, the two successive gene expression times of 23 hrs and 01 hr respectively are separated by 2 hrs and not 22 hrs, and in terms of equivalent angular representation, the mean angle is 0 deg (corresponding to midnight 24 hrs as the origin 0 deg) and not 180 deg. Directional statistics deal with rigorous and proper methods for analysis of angular data and of strictly periodic events, since the latter can be mapped onto the circumference of a unit circle. First, constructions of probability distributions on the non-Euclidean product manifolds of Torus ($\mathrm{S^1xS^1}$), Cylinder ($\mathrm{R^1xS^1}$), Hyper-Disc ($\mathrm{R^1xS^2}$), etc. are presented. The indispensability of these distributions in proper statistical inference and also in Statistical Machine Learning for manifold data, along with some hurdles, are presented. Finally, several emerging real-life examples from biological sciences and bioinformatics are given to illustrate the proposed theories and methodologies.
For further information, contact Farhad Shokoohi.
