Privacy and Game Theory
I’m working to model the behaviors of agents who prefer not to reveal their preferences to a mechanism (in the context of algorithmic mechanism design). We find that differentially private mechanisms may reduce net social utility.
Training Discrete VAEs for Sentences
Some methods to improve training of discrete variational autoencoders applied to natural language processing. Specifically, we use clustering algorithms to improve the accuracy of biased gradient updates.
Heterogenous Effects in Literacy Interventions
We identified strong heterogenous effects from a randomized control trial studying early language literacy interventions in Miami-Dade.
A geospatial package I wrote at the Manhattan Institute for a report on urban Texas housing markets. It’s intended to be a fully reproducible data project.