Shanshan Tu, PhD graduate in Statistics from Ohio State University
I am interested in utilizing statistics, machine learning, and natural language processing techniques to solve real-life problems. My research focuses on the robustness and model complexity in support vector machine (SVM), a popular margin-based classifier. In general, due to the importance of model stability for prediction performance, people are interested in how sensititve the margin-based classifier is to some change of the data. For instance, in least squares regression, case influence is extensively studied for model diagnostics (e.g. Cook's distance), model selection (e.g. Leave-one-out CV), and model complexity (e.g. Model df). I will talk about how to generalize the idea to assess case influence in classification, specifically in SVM. I will focus on an efficient homotopy computational strategy that we proposed to assess case influence in SVM in terms of case deletion and label flipping.
Social hour following: 4-5pm