Event Prediction Research Wins AI and Statistics (AISTATS) Conference Honor
Researchers at UC Irvine’s Donald Bren School of Information and Computer Sciences (ICS) have been honored by the 29th Annual Conference on Artificial Intelligence and Statistics (AISTATS 2026) for a novel machine learning approach called EventFlow,
which could help predict future events over time more simply and accurately.
Padhraic Smyth, Distinguished Professor of Computer Science and Hasso Plattner Endowed Chair in Artificial Intelligence at ICS; Gavin Kerrigan, Ph.D. ’24 alumnus in computer science, currently a postdoctoral research assistant at University of Oxford; and Kai Nelson, former ICS visiting researcher in computer science, now a Ph.D. student at UC Berkeley, won AISTATS Best Paper Award for their paper titled “EventFlow: Forecasting Temporal Point Processes with Flow Matching.”
The ICS researchers won the honor amid over 2,100 submissions to the AISTATS conference, which focuses on the intersection of AI and statistical methods and is considered one of the top international research conferences in artificial intelligence.
The paper proposes a new machine learning approach for prediction of event data over time, using a combination of deep learning and a technique called flow matching. With its simple implementation and 20-53 percent lower forecast error than standard
benchmarks, its impact is potentially wide reaching.
Smyth explained, “In this exciting new work, led by Gavin and Kai, we developed a novel machine learning approach called EventFlow for prediction of events in time – the method is quite different conceptually to existing models and algorithms and produces more accurate predictions than the prior state-of-the-art. EventFlow has broad potential impact across a range of real-world applications, including better prediction of health outcomes in medicine, more accurate financial forecasting, understanding analytics
data in education, predicting wildfire ignitions, and analyzing online human behavior data.”
“Our research tackles a question that is simple to state but hard to answer: How can we predict when an event will happen?” Kerrigan said. “By learning from patterns in past data, we aim to forecast when future events are likely to occur. Whether it’s aftershocks
following an earthquake, patients arriving at a hospital, or patterns of brain activity, accurately predicting these timings is critical. We introduce a new approach, inspired by recent advances in generative modeling, that significantly improves our ability to make these predictions and could have a broad impact across these areas.”
Kerrigan looks forward to attending and presenting at the AISTATS conference being held on May 2-5 in Tangier, Morocco. He added, “I’m incredibly honored to receive this award, and very grateful to my collaborators, Kai and Padhraic, for making this work possible.”
The research was funded by a combination of funding from the National Science Foundation, the National Institutes of Health, and the Hasso Plattner Institute (HPI) Research Center in Machine Learning and Data Science at UC Irvine.
– Tonya Becerra