Models are tools that scientists use to make sense of data, explore new ideas, and generate new avenues of research. Some models are physical or biological, like the solar system model of the atom or the biologist's trusty fruit fly. But many scientific models take the form of a set of equations or algorithms that can be implemented as computer programs. Such computational models can be extremely useful in deciding between competing theories and ideas, because their corresponding computer programs can be pitted against each other to see which one provides a better account of the data. The problem is that often it is not so easy to determine what a "better" account is. A model might capture one data set very well, but provide little insight beyond that particular data set. Since scientific theories are usually concerned with general principles rather particular sets of data, it is useful to have methods that are sensitive to more than just the fit of a model to a given data set.
With support of the National Science Foundation, Dr. Thomas will develop methods that can be used broadly to test models against one another on the basis of model form and function, rather than just fit to data. These methods will be developed in the context of theories of human pattern recognition, a topic that is general enough to be relevant to range of areas in the behavioral and social sciences. From hearing a word to driving a car to making an investment, the ability to detect and recognize patterns in the world is central to human activity. The importance of this topic has led many researchers to formulate many computational models that capture theories about how patterns are recognized in the brain and mind. Therefore this topic domain provides an apt testing ground for the development of methods that will help to identify theories that are and are not worth pursuing as evidence comes to light.
