Explainable AI-Driven Early Microlensing Discovery in the Roman GBTDS
| Program ID | 19058 |
|---|---|
| Science Category | Exoplanets & Exoplanet Formation |
| Program Type | Analysis |
| Category | Small |
| Principal Investigator | Javier Viaña |
| PI Institution | Massachusetts Institute of Technology |
| Co-Investigators |
|
| Abstract | Roman’s Galactic Bulge survey will detect thousands of microlensing events, but many will peak outside the observing window due to seasonal gaps. This jeopardizes our ability to characterize black hole candidates, high magnification events and planetary anomalies. To solve this, we propose the development of an AI-based pipeline for early detection of microlensing events, while they are still rising, so that follow-up observations can capture their most important phases. The core of the system is RomanLensNet, an explainable neural network trained to recognize microlensing signals from pre-peak light curves. Thanks to the explainability component, each prediction will be accompanied by a a reasoning map, which will help us to understand and validate the model's decisions. Candidate events are evaluated, modeled to estimate time to peak, and prioritized for monitoring. Finally, we will disseminate the alerts to the community via a public website, a Slack channel, and an email list. This project will maximize the impact of Roman, allowing the characterization of events that would otherwise only be partially observed. |