View Proposal #237
If this proposal belongs to you, you are authorized to update it. Use the menu on the right.
ID | 237 |
---|---|
First Name | Fengrong |
Last Name | Wei |
Institution | University of Iowa |
Speaker Category | graduate student |
Title of Talk | variable selection in high dimensional regression |
Abstract | My research work studies statistical regression models for data sets with a small sample but huge number of variables. For example, we may wish to study the same 5000 genes in only 200 individuals with the goal of predicting whether they will develop a certain rare cancer. A classical linear regression for the cancer outcome in terms of the 5000 genes does not work with only 200 data points because the associated linear equations are not full rank. We might choose 200 of the genes and do a regression, but there are over 10^363 such choices. My work uses "penalty functions" add to the linear equations which will make the problem solvable. Theoretically, we can show that the result have the "oracle" property which means it will give us the baseline true model with probability going to 1. |
Subject area(s) | biomathematics |
Suitable for undergraduates? | Yes |
Day Preference | |
Computer Needed? | Y |
Bringing a laptop? | Y |
Overhead Needed? | Y |
Software requests | |
Special Needs | |
Date Submitted | 3/27/2008 |
Year | 2008 |