Course No.：21620Z
Period：10
Credits：0.5
Course Category：Lecture
Aims & Requirements:
“Large p, small n” data refers to data with number of covariates larger than or comparable to the sample size. Such data naturally arise in genetics, genomics, astronomy, economics, and many other areas. In this lecture series, we will discuss new development in regularized estimation and variable selection methodologies, as well as their applications in cancer genomics.
Primary Coverage：
Lecture 1: Cancer genomics and regularized variable selection
• Cancer gene profiling studies: background
• Introduction to statistical methodologies for cancer genomic data
• Regularized variable selection/estimation: from AIC/BIC to ridge to Lasso
Lecture 2: Bridge penalized variable selection/estimation
• From Lasso to bridge
• A fast computational algorithm
• Asymptotic properties of bridge estimates
• Applications in cancer genomics
Lecture 3: Incorporating prior information in bridge estimation
• “Group” penalization: from group Lasso to group bridge
• The coordinate descent algorithm
• Applications in pathway analysis of cancer genomic data
Lecture 4: Threshold gradient directed regularization: a new set of regularization
• The TGDR algorithm
• Comparisons with bridge penalization: pros and cons
• Applications in cancer genomics
Lecture 5: Incorporating prior information in TGDR estimation
• Cov-TGDR: incorporating multiple sets of covariates
• CTGDR: incorporating the clustering structure
• MTGDR: incorporating multiple heterogeneous datasets
At end of the lecture series, audiences are expected to have a good understanding of the frontiers and open questions of methodologies for “large p, small n” data, as well as basic understanding of cancer genomics.
Author：Shuangge Ma
