March 22, 2019
Colloquium Topic: Predicting the Future: IARPA Geopolitical Forecasting Challenge and Crystal Cube
The talk will describe methods for predicting the future, developed by a team of APLers. Team DigitalDelphi earned first place overall and Election Forecaster honors in the Intelligence Advanced Research Projects Activity (IARPA) Geopolitical Forecasting Challenge. IARPA created the challenge as a way for the intelligence community to develop innovative ways to use crowdsourced forecasts and other data to predict potentially disruptive geopolitical events.
For seven months, competing teams were tasked to submit predictions for 165 questions on various geopolitical topics, such as whether any NATO member would invoke Article 4 or Article 5 of the North Atlantic Treaty, whether Afghanistan's president would experience a significant leadership disruption, or how many Middle Eastern countries might return their ambassador to Qatar. For each open question — sometimes there were 80 questions open at the same time — a forecast needed to be submitted every day
The second part of the talk will concentrate on a Propulsion Grant called Crystal Cube. The goal of Crystal Cube is to create an automated capability for the prediction of disruptive events. The disruptive events we want to predict are wide-ranging and include armed conflict, insurgency, overthrow of dictators, economic collapse, failed states, and novel attacks on the US and other countries. Artificial Intelligence methods such a deep learning, random forests, and clustering are used for forecasting of disruptive events.
Colloquium Speaker: Anna L. Buczak
Dr. Anna L. Buczak is the Chief Scientist of Systems Integration Branch in AMDS. She holds a PhD in Applied Artificial Intelligence and an MS and a BS in Computer Science.
Dr. Buczak has served as Program Manager or Principal Investigator on multiple R&D projects, dealing with disruptive events forecasting, disease prediction, pattern recognition, anomaly detection, system self-organization, and intelligent agent learning. Her research has been supported by DARPA, CDC, DoD GEIS, USAMRMC, and IARPA. She is an Associate Editor of BMC Medical Informatics and Decision Making. From 1998 to 2007 she has been a Co-Chair Artificial Neural Networks in Engineering Conference; she has authored and coauthored over 80 research papers and holds six US and international patents. She has been a Ph.D. dissertation co-adviser and a member on Ph.D. Committees at Rutgers University. Presently she is a Doctor of Engineering Advisor in the JHU DoE Program. She is a member of IEEE.
Dr. Buczak’s research interests include machine learning with a special emphasis on biologically inspired learning methods, anomaly detection, data mining, and self-organizing and adaptive systems. The novel methods she develops perform disruptive events prediction, detection of cyber attacks, disease outbreak prediction (dengue, malaria, flu), and detect anomalies in various data streams.