Lopata Hall, Room 101
"Decoding Cellular Circuits with Matrix Factorization"
Adviser: Michael Brent
Environmental changes trigger the circuitry within cells to respond with changes in the amount of gene product created. This circuitry of cells comes in multiple layers, and involves connections with factors including outside environmental signals and internal feedback signals. Unfortunately, most of a cell's internal state cannot be measured directly due to financial constraints, time constraints, or just insufficiently advanced measurement technology, so a major area of research in systems biology is reverse engineering this information from the limited kinds of gene products that can be efficiently measured. This talk will show how matrix factorization can be used in this problem, by structuring a model for inferring the internal state of a cell's circuitry.
"Using Text-Based Measures of Ideology to Analyze Political Communication"
Adviser: Sanmay Das
With the rapid developments in machine learning for text classification and categorization, there has been much interest in developing and using text-based measures of everything from sentiment to political ideology. Much of this research begs the question of whether the target remains consistent across different types of text. We investigate this question in the context of political ideology. We use text data from different sources related to US politics (namely Wikipedia, the Congressional Record, and Press Releases from members of Congress) to measure ideology. We find that, while it is relatively easy to predict partisanship from text, it is much more difficult to consistently measure political ideology within party. The problem is particularly egregious in text that is created by the politicians themselves; indeed, we provide evidence that congresspeople may use different communication channels to express significantly different ideological leanings.