Events

Seminar: “A Framework for Solving Complex Biomedical Problems”


Printer friendly page

Marianthi Markatou
Columbia University

December 3, 2007
2:30-3:30 PM
341 Coverdell Center

ABSTRACT

 

Scientific advances have presented statistical sciences with many interesting challenges, one of which is the challenge of large dimensions, both in terms of the number of data vectors and of the dimension of the data vectors. Assessment of the fit of a model in such situations may be challenging.

We will discuss a framework for high dimensional model assessment that has the potential to be useful in solving complex biomedical problems. This framework is based on the concept of distances and has connections with the usual goodness-of-fit literature.

What distances are useful for high dimensional data? We answer this question by considering a wide and flexible class of distances that is characterized by a simple, quadratic form structure that is adaptable through the choice of a non-negative definite kernel. We illustrate connections with the field of machine learning and discuss potential applications in the field of genomics where data may be represented as long binary vectors, and in clinical medicine, thereby indicating a vision for research and training in biostatistics.