Ashketar Frontiers of Science Lectures Presents Helen Greatrex and John Harlim скачать в хорошем качестве

Ashketar Frontiers of Science Lectures Presents Helen Greatrex and John Harlim

"Forecasting Without Data: Rethinking Weather Observations for Disaster Response" Helen Greatrex, assistant professor of geography and statistics We’re used to weather apps disagreeing, but in disaster response, choosing the wrong forecast can have real consequences. Normally, models are validated using weather station data, but in much of the world that data is limited or missing. This talk explores how Penn State researchers are finding creative ways to work around these gaps. From using newspaper archives to reconstruct flood histories in India, to combining humanitarian narratives with satellite rainfall estimates in Somalia, to co-designing drought insurance tools with farmers, the talk will look at how unconventional data sources can make forecasts more useful and trustworthy where they’re needed most. "The role of mathematics in modeling and prediction in the era of AI" John Harlim, professor of mathematics The rapid rise of artificial intelligence (AI) has transformed many aspects of daily life, from performing routine tasks such as generating dinner menus to addressing complex scientific problems such as predicting weather. Despite their remarkable empirical success, AI-driven approaches continue to face fundamental challenges related to reliability, interpretability, and long-term predictive fidelity. In this talk, Harlim will illustrate these challenges with examples drawn from aerospace engineering and climate prediction and highlight how his research group contributes to addressing these issues. In particular, Harlim will focus on the basic problem of identifying mathematical representations that enable reliable long-term prediction from noisy observational data. He will demonstrate structure-aware mathematically grounded methods that can either significantly outperform or enhance the prediction skill of AI-based models.