Research with Prof. Joseph K. Blitzstein concerning Respondent-Driven Sampling and Systems, related estimation problems, and goal-based process incentive optimization. Also, work on network visualization and node attribute-dependent network structures such as homophily.
Work on optimization and response surface methods with marketing applications. We consider applied example of a large consumer packaged goods company (CPG) with response surface optimization of dimensionality exceeding 300,000,000. To perform optimization, we use deterministic numerical algorithms to find local modes. Apart from computational aspects, we study the problem's properties from the perspective of modern experiment design, where such high dimensionality and related methods have not been considered so far. Research in collaboration with Tatsunori Hashimoto.
Cross-validation methods are a useful tool for model selection and overfitting detection. This research aims at developing its theoretic foundations and optimality criteria further beyond jackknife and putting cross-validation in context together with other methods such as AIC and BIC, as well as Bayes Factor, which in many circumstances prove to be impractical.
This research direction stems out of my applied interest in interactive visualization as a tool to convey research goals, process, and findings. I have created many visualizations with this purpose, such as those appearing in this talk. Interactive visualization is also a great tool for sensitivity analysis and as an intermediate step in statistical model-building. I am interested in studying how visualization can be used to fit and provide feedback on model nuisance parameters.