Nikhil Devanur Thesis
Leveraging a primal-dual connection, we map features of the propensity score model to choices about the specification of the original SC optimization problem.
In particular, the original SC method, which balances the L2 norm of pre-treatment outcomes, is identical to IPW with an L2 penalty on the propensity score model and with the least feasible regularization; other choices of balance criteria (e.g., L-infinity norm) correspond to other forms of regularization (e.g., Lasso).
Or, how do algorithms circumvent various lower bounds when it comes to real-world graphs?
This question has no simple answer, and I will present a tale of two stories on this theme.
His primary interest is in mathematical foundations of big data, especially modeling and algorithms.
A curious feature of the mathematics is the use of Turan's theorem in extremal combinatorics, to prove correctness of the algorithm.
We are particularly interested in the high-dimensional case when d is large.
We study a basic hypothesis testing problem: can we distinguish a random geometric graph from an Erdos-Renyi random graph (which has no geometry)?
Bio: Avi Feller is an assistant professor at the Goldman School, where he works at the intersection of public policy, data science, and statistics. in Applied Statistics as a Rhodes Scholar at the University of Oxford, and a B. in Political Science and Applied Mathematics from Yale University.
His methodological research centers on learning more from social policy evaluations, especially randomized experiments. Algorithmically exploiting structure or: how I learned to stop worrying and enjoy real-world graphs C.