American Statistical Association
The increasing availability of multi-view data (data measured on the same samples from multiple sources) has generated strong interests in integrative data analysis. Different data views usually contain shared information, while each view may have its unique structure as well. Factorization methods are commonly used to separate shared signals from individual signals preceding further analyses. However, there remain significant challenges, especially in the modeling of 1) partially-shared structure, 2) non-Gaussian data, and 3) block-wise missing data (values for an entire source are missing).
In this talk, I will first introduce a new dimension reduction method, which adaptively learns the structure-sharing patterns and estimates the underlying signals of multi-view data. Then I will present a new low-rank model for non-Gaussian multi-view data with block-wise missing values. The methods can be readily used for pattern recognition and missing data imputation, and the signals estimated from the methods facilitate further inferences such as clustering and prediction. I will demonstrate the advantages of the proposed methods on a breast cancer study from TCGA and a mortality study.
|Date:||Wednesday, December 6, 2017|
|Time:||4:00 - 5:00 P.M.|
Memorial Sloan Kettering Cancer Center
Department of Epidemiology and Biostatistics
485 Lexington Avenue
(Between 46th & 47th Streets)
2nd Floor, Conference Room B
New York, New York
**Outside visitors please email firstname.lastname@example.org for building access.
You must be on the security list to enter the floor.