Spectral Models of Security: New Methods for Detecting Evolving Community Structure in Societal-Scale Data
Principal Investigator: Joshua Blumenstock, University of California Berkeley
Co-Principal Investigators: Sham Kakade, University of Washington; Jacob Shapiro, Princeton University
Years of Award: 2017-2020
Managing Service Agency: Office of Naval Research
We are developing new, scalable methods for modeling and studying evolving, societal-scale communities and social networks. The technical focus of our work is on adapting spectral machine learning methods to real-world contexts where digital network data is very large, sparse, and dynamic. This requires methodological innovation to make existing algorithms better suited to hypothesis testing and to non-stationary processes. The practical application of this method is to study the impact of significant geopolitical events on the social fabric of micro- and meso-scale community networks in multiple developing countries. Specifically, we are analyzing terabytes of anonymized mobile phone data from Afghanistan and Rwanda to understand how violent and non-violent events affect the cohesion, fragmentation, and internal displacement of local ethnic communities.
Cadamuro, Gabriel, Ramya Kolakai Vinayak, Joshua Blumenstock, Sham Kakade, and Jacob Shapiro. 2019. "The Illusion Of Change: Correcting For Biases In Change Inference For Sparse, Societal Scale Data". In WWW '19 The World Wide Web Conference, 2608-2615. New York, NY: Association for Computing Machinery.
Khan, Muhammad Raza, and Joshua E. Blumenstock. 2019. "Multi-GCN: Graph Convolutional Networks For Multi-View Networks, With Applications To Global Poverty". Proceedings Of The AAAI Conference On Artificial Intelligence 33: 606-613.
Raza Khan, Muhammad, and Joshua Blumenstock. 2017. "Predictors Without Borders: Behavioral Modeling Of Product Adoption In Three Developing Countries". In KDD '16, 145-154. New York City: Association or Computing Machinery.
Sharan, Vatsal, Sham Kakade, Percy Liang, and Gregory Valiant. 2019. "Prediction With A Short Memory."