Flexible and Adaptive Statistical Modeling

点击翻译按钮获取中文摘要

In the past 10-15 years there have been significant developments in theareas of applied regression and classification. Much of the impetusoriginally came from outside of the field of statistics, from areassuch as computer science, machine learning and neural networks. In somecases, statisticians have helped to synthesize these innovations, often byreinterpreting them from the point of view of standard statistical models.As a result, we now have at our disposal a very powerful collection oftechniques for adaptive regression and classification. These are nowbeing applied to medical diagnosis, bioinformatics and genetic modeling,chemical process control, shape, handwriting, speech and face recognition,financial modeling, and a wide range of other important practicalproblems. In this proposal, the investigator develops new methods for adaptive regression and classification, with particular applicationto problems in bioinformatics and genomics. One such method is the"fused lasso", a technique which seeks solutions which are both sparseand and smooth. It is designed for problems in which the features havea natural ordering, e.g. in time or space. Protein mass spectometryrepresents one such area of application.In this research, the investigator develops new statistical algorithmsfor analyzing data from biological and human experiments. This workis part of the exploding field of "bioinformatics", which seeks to makesense of the huge volume of data produced by new technologies in biologyand medicine. This kind of work represents the next step after thesequencing of the human genome, and has enormous potential for societalimpact through disease prevention and treatment.