Séminaire Exoplanètes / |
| « Can Convolutional Neural Networks Help Us Identify Gravitational Microlensing Exoplanets? » |
|
Stela Ishitani Silva |
| Gravitational microlensing is one of the primary techniques for detecting exoplanets, being particularly sensitive to low-mass planets that orbit at moderate to large distances from their host stars. This sensitivity is crucial for complementing statistical analyses derived from other exoplanet detection methods, allowing coverage of a broad range of planet-to-host star mass ratios and orbital separations. However, identifying planetary microlensing events within hundreds of thousands of available photometric light curves is challenging. In response, Convolutional neural networks (CNNs) can be a viable alternative for identifying such events, especially in massive datasets, such as the enormous amount of data from the upcoming Nancy Grace Roman Space Telescope. In this presentation, we will highlight our ongoing efforts to implement a CNN-based pipeline for detecting gravitational microlensing events using the 9-year dataset from the Microlensing Observations in Astrophysics collaboration. Building upon our successful application of CNNs for detecting planetary transit signals in data from the Transiting Exoplanet Survey Satellite (Olmschenk, Ishitani Silva, et al 2021 AJ 161 273), we use only raw photometric light curves—without any prior fitting or feature extraction—as input for our CNN. Additionally, I will address the challenges associated with composing a comprehensive training dataset that captures the various shapes of microlensing events. With the ability to infer each light curve in milliseconds, our approach aims to accelerate the identification of microlensing exoplanets. |
|
mardi 4 juin 2024 - 14:00 Salle des séminaires Évry Schatzman Institut d'Astrophysique de Paris |
| Page web du séminaire / Seminar's webpage |