UM Computer Science Professor Jesse Johnson is working with colleagues in UM Geoscience to develop an ice sheet model capable of combining the best available observations with detailed physical descriptions of the ice. Once the data has been 'assimilated', the model can be used to asses the impact of climate change on the ice sheet's stability. Model output in this figure of Greenland shows the results of a recent assimilation of satellite derived surface velocity data into the model. The unstructured mesh (black triangles) delivers resolution where it is needed, but reduces computational overhead in regions of slower change. Ultimately, the purpose of models like Johnson's is to determine the amount of sea level change that ice sheets produce.
This research uses highly parallel computation to implement a set of algorithms which allow hundreds of gigabytes of text based data to be categorized based on semantic meaning, regardless of the specific words or phrases used. The research product works on a variety of character based languages and has even been able to analyze and categorize mixed language text (i.e. Spanglish). Rather than publishing the results, Dr. Henry has licensed the research from UM for commercialization through his company, Agile Legal Technology. Subsequent work done through this company has led to a patent application as well as development of a cloud based solution.
In collaboration with partners in the Biochemestry Department, several of the graduate students and Dr. Raiford will begin investigating methods of identifying common secondary (and possibly tertiary) structures of functionally similar RNA sequences. This will involve the analysis of experimentally derived data on randomly generated RNA sequences and their affinity for a particular protein (nucleocapsid protein (N) from River Valley Fever Virus (RVFV)). All of the resulting RNA sequences will have a known affinity for this protein (aptamer status) and it will be our job to identify their secondary structures as well as common structural features, the implication being that these secondary structures (and their resultant tertiary structures) are responsible for the sequence's high affinity for the target protein. Algorithmic approaches will include calculation of tree edit distances, dynamic programming techniques for identifying minimum energy states, as well as clustering and classifier algorithms.
Genetic algorithms are a family of computational methods inspired by evolution. Potential solutions to a problem are encoded in a chromosome-like data structure, and then simulated evolution is applied to a population of these chromosomes. Variational operators like mutation and recombination are applied to the chromosomes, and chromosomes corresponding to good solutions are favored during reproduction. Genetic algorithms have been successfully applied to solve many kinds of problems. In order to better understand how genetic algorithms work, they have been mathematically modeled as dynamical systems. For some special kinds of population (no mutation and a population consisting of only one kind of individual), the population is a fixed point of the dynamical system. A fixed point is stable if points near the fixed point converge to the fixed point under repeated iterations of the dynamical system. The stability of these no-mutation fixed points can be determined from previous research, but it was not known how the fixed point changed as mutation was introduced. I and my coauthors have mathematically proven that stable fixed points remain stable while unstable fixed points "disappear" as mutation is introduced into the system. This confirms our intuition and helps our understanding of the relationship between mutation and recombination in evolutionary systems.
Department of Computer Science
University of Montana | Social Sciences Bldg. Room 401 | Missoula, MT 59812
Department Chair: Yolanda Reimer | (406) 243-4618 | firstname.lastname@example.org
Office Contact: Robyn Berg | (406) 243-2883 | email@example.com