By combining high-resolution smartphone location data with data from population census, we estimated the heterogeneity of the response to non-pharmaceutical interventions across socioeconomic groups in middle income countries, finding that high-wealth smartphone users consistently reduced their mobility about twice as much as low-wealth users. These results suggest that developing-country governments should instead carefully weigh the costs and benefits of social distancing policies especially on the most vulnerable populations. KDD, August 2020.

We show that training state-of-the-art language models using active learning can help predict Twitter users' unemployment status and variations in unemployment statistics both in the US and in middle income countries. IC2S2, July 2020.

Predicting risks of famine and food insecurity using natural language processing @ World Bank Econothon, July 2019.

We often think of performance and success interchangeably. However, in areas of human activity where performance is difficult to quantify in an objective fashion, reputation and networks of influence play a key role in determining who becomes successful. In this talk, I present some of my research that shows how big data and network science can help us predict artistic success @ TEDx, April 2019.