Grant: Discovery Grants Program - Individual

My research focuses on Bayesian networks, which are a formal framework for uncertainty management based on probability theory. In particular, I study Bayesian network inference. Our research program may lead to a better understanding of the theory of evolution. The field of phylogenetics presents difficult statistical and computational problems. Given the flood of molecular data, it is not feasible to test all possible evolutionary trees for some sample data. Instead, only a portion of possible trees are scored. By providing a more informative and efficient method for scoring trees, our approach will score more trees in the same amount of time, likely including trees that more accurately reflect the evolution of species. Hence, it is reasonable to expect that our system, called DataBayes, will provide better results, both in terms of accuracy and efficiency, than MrBayes, a standard program for the Bayesian estimation of phylogeny. By providing a better evolutionary tree, the impact of our research findings may be immense. Howe, a collaborator from Cambridge University, shows that many other areas follow a similar phylogenetic method of DNA copying with the incorporation of changes, including manuscript versions of texts, languages, musical instruments, commercial organizations and Persian carpets. Therefore this research proposal can have far reaching effects. Semantics of inference may be exploitable in the work on fusing Bayesian knowledge sources done at Dartmouth, and research conducted at MIT on teaching a computer to play a complicated game like Civilization. Similar to the problem of scoring evolutionary trees above, it is not possible to consider all instances of Civilization. However, semantics may allow for more instances to be considered, thereby leading to better computer actions.