Embedded Files
🤖 CLASSIFYING EXOPLANET TRANSITS WITH CNNs 🤖
🤖 CLASSIFYING EXOPLANET TRANSITS WITH CNNs 🤖
In 2018, I participated in the NASA Frontier Development Lab, which pairs machine learning experts with science researchers to tackle challenges in the space sciences of interest to NASA. My team focused on classifying candidate exoplanet transits detected by the Kepler mission using deep convolutional neural networks (CNNs). Below is a short video by Google Cloud that tells the story, and also my talk at Google Cloud's Next '19 event in San Francisco where I highlighted the role of the Google Cloud Platform in our research. You can find more details on our work in this ApJ Letter or a brief summary in this AAS Nova article.
đź’Ą DETECTING FLARES IN TESS DATA W/CNNs đź’Ą
đź’Ą DETECTING FLARES IN TESS DATA W/CNNs đź’Ą
I've also been working with with Adina Feinstein, a PhD student at the University of Chicago, to design a CNN that automatically and rapidly detects stellar flares in light curves from the Transiting Exoplanet Survey Satellite (TESS). The CNN, which is affectionately named Stella, is really good at it's job---it achieves high (99%!) precision and with minimal data preprocessing (no de-trending of stellar activity signals needed!). The animation below shows how we can feed TESS light curves into this trained CNN “classifier” to transform it into a “detector” that outputs a “flare probability time series” (where I use the term “probability” very, very loosely here). Stella is also fast---it takes roughly 30 minutes to search 3500 TESS light curves for flares. The paper will be out soon, but for now, we made stickers out of the sweet Stella logo that Adina designed (see below), so ask us for one.
Google Sites
Report abuse