Half year program proposal for 2004-2005    (PRELIMINARY DRAFT)

o Scientific description

We take as motivation for this proposal, a recent MSRI workshop entitled ``Nonlinear estimation and classification." The goal of that meeting was to bring together researchers from various disciplines, each facing the formidable task of analyzing massive amounts of high-dimensional and highly-structured data. Due in part to recent advances in data collection and computing technologies, many fields are now grappling with statistical questions relating to large, complex data. Each such problem area is a challenge to the field of statistics, a challenge to remain relevant.

The last MSRI program in statistics was held during the 1991-1992 academic year. The program was divided between a semester on empirical processes theory and semiparametric models, and a semester on computer-intensive methods. The organizers placed statistical theory and general methodology at the center of their program. In 2001, however, an amazing array of applied areas have been pushing the boundaries of statistics and are providing fruitful areas for fundamental statistics research. As a result, substantial inter-disciplinary collaborations are becoming mainstream.

As a reflection of this new emphasis, we propose a semester-long program in statistics that is organized around two broad, inter-disciplinary research areas. As with the ``Nonlinear estimation and classification" workshop, our program will involve eminent experts from the various fields in order to exchange ideas and forge directions for the future. In part, this program is an attempt to highlight important new statistical problems, many of which have emerged in other disciplines. We hope to establish some consensus within the statistics community about new research programmes. Given this forward-looking position, we expect our proposed statistics year to have the most impact on new researchers. Hence the bulk of our funding will go to support the participation of students and new PhD's. We plan to have a series of workshops to target the subject areas listed below. The areas of the long-term participants will be coordinated with the workshop series and we will balance the proportion of senior and junior researchers at any given time in residence.

Problem Areas:

  Information technologies

Information Technologies (IT) are changing every facet of our lives. At the heart of this IT revolution are the scientific and engineering problems which are statistical in nature and they provide a fertile ground for application of statistical methods and fundamental statistical research. These areas include data compression, signal processing, fMRI and medical imaging, Internet data, wireless net, communications, digital libraries, and statistics of multimedia. They stand out because of the massive amounts of data collected and the common requirement of processing in an on-line fashion. Computation efficiency and statistical estimation efficiency have never been in more close competition than in these problems. This IT problem solving requires collaborations of statisticians, computer scientists and electrical engineers. The proposed MSRI year will have the fall semester focused on bringing together researchers from these different disciplines as long term residents and also via two workshops.

  Bioinformatics/Statistical Genetics

The Human Genome Project is entering its second phase, functional genomics, to understand the biological meaning of DNA sequences. At the same time, the identification of genes contributing to complex human diseases has become an important and, it is hoped achievable objective. Part of our MSRI program will be built around two main themes: the analysis of gene expression data and statistical linkage analysis for complex traits.

DNA microarrays and other ways of collecting data on the expression of thousands of genes simultaneously permit biologists to address many interesting problems including tumor classification and gene identification. At the same time this technology poses many challenging statistical problems, especially in the area of clustering and classification.

Linkage analysis activities will be organized around methods for the use of data on pedigrees (closely related individuals among whom relationships are known) and the use of data from structured populations, that is, more remotely related individuals, among whom precise relationships are not known.

Many of the problem areas listed routinely deal with massive data, often requiring fast or real-time processing. These areas also overlap with many other mathematical sciences including optimization, computer science, AI/machine learning, signal processing/information theory, and computational biology.

o Program Organizers (TBD)

David Donoho* (Stanford, Statistics)
Mark Hansen (Bell Laboratories, Statistics)
Deborah Nolan (UC Berkeley, Statistics)
Terry Speed (UC Berkeley, Statistics and Biostatistics)
Elizabeth Thompson (University of Washington, Statistics and Biostatistics)
Martin Wells (Cornell, Social Statistics)
Bin Yu (UC Berkeley, Statistics)

* David Donoho is also helping with a proposed program on image analysis with David Mumford. If this program is scheduled during our year-long program, then David Donoho has agreed to be a co-organizer for both, working to maximize the synergy between the two programs.

It is clear that the changing face of statistical research should also influence how we train our students. In parallel workshops, we plan to address the teaching of statistics in two ways. First, we intend to explore the impact of collaborative work on the design of statistics curricula. From computing to methodology and theory, this program will highlight a number of new directions for how we train the next generation of statisticians. Second, we will explore how new technologies will facilitate the act of teaching statistics.