AI for Scientific Data Streams: Real-time detection of change and anomalous structure

A steel plant’s X-ray scanner, a particle accelerator’s detectors, and a weather satellite’s sensors all produce torrents of data that can hide a crucial glitch or a new discovery in plain sight—and this project will build software to spot those signals the instant they appear. Current methods for detecting when a scientific instrument starts recording something unusual—a structural change in a material, a rare particle event, or an unexpected shift in a sensor reading—often require scientists to stop and manually inspect the data, or to rely on bespoke algorithms that work for only one type of experiment. This project will develop general-purpose, real-time anomaly detection tools that can be dropped into any scientific data stream, from materials science to frontier physics. The researchers will work directly with leading experimental groups to test and refine the methods on real problems. If successful, the tools could accelerate discovery across multiple fields—for example, flagging a new material phase as it forms inside a synchrotron beamline, or catching a transient signal in a physics detector before it is buried in noise. The project is primarily about building fundamental computational methods, not a specific application, but the same techniques could eventually find use in monitoring industrial manufacturing lines, energy grids, or environmental sensor networks for early signs of failure or change.