# MIT's Machine Learning Algorithm Generates Extreme Event Scenarios A new method models future extreme events without relying on historical data. - **Source:** [MIT News — AI](https://news.mit.edu/topic/artificial-intelligence2) - **Published:** 2026-08-24 - **Topic:** [Data & AI](https://brifing.si/c/data-ai) - **Original article:** - **On Brifing:** https://brifing.si/s/57f69219-5945-48d4-a1e6-77768d47d0e8 ## The Challenge of Predicting Extreme Events Predicting rare events like extreme weather, wildfires, or financial crises is difficult with traditional methods because these events are poorly represented in historical records. MIT engineers developed a machine learning algorithm that can generate plausible scenarios for events not present in past data. The method learns from available data, such as a region's daily weather records, to calculate probabilities for unprecedented extreme events. ## Fundamentals of η-Learning The Extreme Event Aware (η-learning) method statistically learns relationships between two types of data: point statistics and spatial maps. For example, working with precipitation maps of the US, the algorithm can predict high-resolution rainfall patterns from low-resolution inputs. Remarkably, it can generate realistic scenarios for extreme rainfall events not present in the training data. ## Modeling 100-Year Storms The algorithm can map the probable intensity, duration, and impact area of storms that might occur once every 100 years in a region. For instance, while New York's highest recorded rainfall is 200mm, the model can simulate what a 300mm rainfall scenario would look like. Such projections provide critical information for urban planners strengthening infrastructure. ## Applications in Financial Markets The method isn't limited to weather events. It can also model complex extreme events like financial market crashes. By analyzing interactions between different sectors, the algorithm identifies scenarios that could lead to market collapses, offering valuable tools for institutions developing risk management strategies. ## Overcoming Training Data Limitations Traditional methods require records of past extreme events to make predictions. MIT's new algorithm can make accurate predictions using just the first six months of data, which might not contain any extreme rainfall records. This enables effective risk assessment even in regions with data constraints. ## Open-Access Scientific Publication The research team published their method in an open-access paper in Nature Communications in August 2023. The paper shares technical details of the algorithm and its successful applications on US precipitation maps. This publication establishes a foundation for the scientific community to improve and apply the method across different domains. ## Practical Benefits for Decision Makers The algorithm assists municipal authorities, insurance companies, and policymakers in making critical decisions. For example, it helps prioritize infrastructure investments to reduce flood risks in specific areas. It also serves as a valuable tool for financial regulators to test market collapse scenarios. ## Future Application Areas Researchers note the method could be applied to other fields like robotic navigation. For instance, it could help autonomous vehicles prepare for rare but dangerous traffic scenarios. This wide range of potential applications demonstrates η-learning's promise as an interdisciplinary tool for extreme event modeling. ## Sources covering this story - https://news.mit.edu/2026/generating-scenarios-extreme-events-without-extreme-data-0824 — --- _Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._