About MeI am an Assistant Professor of Econometrics and Statistics at the University of Chicago, Booth School of Business. I received my Ph.D. (2016) and MA (2013) in Statistics, and my Master's in Computer Science (SM, 2011), both from Harvard University. My Ph.D. advisors were Edo Airoldi, David Parkes and Don Rubin, and David Parkes was also my advisor for my CS degree.
My research interests lie at the union of stochastic optimization and causal inference, with applications in complex systems. In particular, I study causal inference and experimental design in complex systems, such as multi-agent economies and social networks, in order to evaluate the efficacy of interventions on such systems. Applications of this research include market design and policy analysis, and present three distinct problems, namely interference, entanglement, and dynamics, each of which can invalidate classical causal methods. I am also interested in the interface of statistics and optimization, particularly in making principled (and possibly causal) statistical inferences from large data sets using implicit stochastic approximation methods, which are numerically stable. For instance, in recent work we derived the asymptotic variance of estimators based on stochastic gradient descent. This result allows us to combine the computational efficiency of stochastic gradient methods with principled statistical methodologies, such as hypothesis testing, confidence intervals, or convergence diagnostics.
- Panos Toulis, "A useful pivotal quantity", (American Statistician, 2016, forthcoming)
- Panos Toulis, David Parkes, "Long-term causal effects via behavioral game theory", (NIPS, 2016, forthcoming)
- Panos Toulis, Edoardo M. Airoldi, "Asymptotic and finite-sample properties of estimators based on stochastic gradients", (Annals of Statistics, 2016, forthcoming)
- Panos Toulis, Dustin Tran, Edoardo M. Airoldi, "Towards stability and optimality in stochastic gradient descent", (AISTATS, 2016)