Computational and Statistical Methods
Computational and statistical methods are the backbone of our work, enabling us to both probe the behavior of complex systems and infer their characteristics (often from fragmentary and error-prone observational data). An important thrust of our work thus involves novel applications of simulation techniques, graph algorithms, machine learning, and computational and Bayesian statistical methods to social, biological, physical, and technical systems with complex structure and dynamics. We also, however, develop novel methodology for both computing and data analysis, including algorithms for stochastic simulation as well as techniques for measurement and analysis of relational data. A major theme of our work in this area is the development of models for relational systems that capture dependence in a mechanistically realistic way, while still being statistically tractable (and usable with actually existing data). We also work on problems of measurement, including sampling designs for dynamic and relational data, models for inference from data sources with complex error structure (ranging from physical instruments to human informants), and novel ways of leveraging spectral or other features to infer latent behavior.
Suggested Publications from NCASD