Generating scenarios for extreme events, without extreme data

Generating scenarios for extreme events, without extreme data

Engineers at MIT have developed a machine learning algorithm called “Extreme Event Aware” (or η-learning), which is used to generate tools that can simulate future extreme events and worst-case scenarios without the need for extreme historical data. This method involves statistically analyzing daily weather records and map data to exclude unreasonable scenarios and identify extreme storms, heatwaves, or wildfires with specific occurrence frequencies (e.g., once every 100 years). It includes information on their duration, intensity, and impact range. MIT graduate student Kai Chang said that this tool aims to quantify unforeseen extreme events that may occur but have not yet appeared in the data, making them more severe than any previous events, thereby helping planners prepare. The findings have been published in the journal *Nature Communications*. Additionally, this algorithm can be applied to other fields such as robot navigation and financial markets to analyze the interaction mechanisms behind complex extreme events that lead to market crashes.