Automated Discovery of Cancer Pathway Models from Biomedical Text
The goal of this project is to revolutionize cancer biology (and discovery science in general) by developing capabilities that will allow to (i) automatically read biomedical literature on cancer pathways (ii) extract causal fragments from the literature (iii) assemble those fragments into a full causal model of a cancer pathway. BM is a highly interdisciplinary project that involves ideas and people from various disciplines – cancer & system biology, machine learning, natural language processing (NLP), among others.
Forecasting Significant Geopolitical Events
Real-world events can often be viewed as salient outcomes of interactions between sets of latent spatiotemporal processes that reflect behaviors of individuals and of populations, and the evolution of environmental factors. Evolution of unemployment rates, economic expansion (or shrinkage), evolution of popular and individual sentiment, increasing radicalization (or marginalization) of demographic segments, health/disease trends, and climate change are all examples of influential latent processes that interact with and shape each other. The goal of the SAFE project is to develop a suite of methods for forecasting significant events in MENA region based automated analysis of various data sources.
Anticipating Large-Scale Cybersecurity Breaches
Addressing rapidly growing cyber threats posed by a variety of state and non-state actors is of paramount importance. EFFECT (Effectively Forecasting Evolving Cyber Threats)seeks to develop technologies that counter these threats by anticipating them before they manifest in the form of an actual cyberattack. In particular, our objective is to develop an end-to-end prototype that accurately forecasts emerging cyber threats (including source, target, attack time, and vector of vulnerabilities) by integrating information from a variety of novel, unconventional sensors within a robust inference framework.
The Big K: Automatically Discovering the Most Informative Factors in Social Phenomena
Human behavior and decision-making is affected by myriad of diverse and overlapping factors. Social science research has generally been limited to small studies that attempt to isolate and test the effects of individual, human- hypothesized factors. The emergence of large-scale data about human behavior based on ubiquitous sensors and computing could allow us to automatically disentangle diverse factors behind human behavior in a principled, data-driven way. The main objective of the BigK project is to develop efficient methods for discovering hidden factors that are as “informative” as possible, while not committing to any modeling assumptions about how the data is generated.