PhD students win the Geopolitical Forecasting Challenge
Can you create a method to predict future geopolitical outcomes that is better than current state-of-the-art methods? Naman Goel and Diego Antognini, two of our PhD students did just that and ranked among the top 5 in different categories of the Geopolitical Forecasting (GF) Challenge.
Organized over the course of seven months, the Geopolitical Forecasting (GF) Challenge was about aggregating information to make accurate forecasts on world events such as political elections, disease outbreaks, and macro-economic indicators.
The IC School was represented by PhD students Naman Goel and Diego Antognini, under the guidance of the Professor Boi Faltings, Director of the Artificial Intelligence Laboratory (LIA) at the EPFL. Two master students, Jean-Thomas Furrer and Jiaxi Gu, also contributed during their semester projects by writing the first implementations of their winning method.
“This competition was a very good opportunity for us to test some of our research in practice and identify new problems to work on”, said Naman Goel.
Combining predictions with dynamic weighing
The LIA team worked on two approaches, one using natural language analysis and other combining crowd predictions with dynamic weighing. The dynamic weighing method was the global winner in the European macro-economics category, and ranked among the top 5 in other categories. Naman Goel and Diego Antognini presented their winning method at an event which took place in Washington DC.
“The information we find on the internet has many contradictions. By using game theory, AI can reconcile them to find the truth, but real scenarios require more than theory. Competitions like this one are very helpful for driving our research forward”, explained Faltings.
The GF Challenge is a global competition run by the Intelligence Advanced Research Projects Activity (IARPA). It involves companies and universities in the US that are funded by the agency, and is also open to participants from around the world. IARPA is an American research intelligence funding agency, working to identify methods to maximize the quality of forecasts that decision makers rely on.