NASA's satellite dedicated to finding exoplanets, TESS (Transiting Exoplanet Survey Satellite),
translates their universe-wide search into a complex data challenge. Every 27 days, TESS
scans a portion of the sky, capturing millions of tiny fluctuations of light as exoplanets
orbit their suns, generatating terabytes upon terabytes of raw data. For researchers, combing
through it all was quite the challenge.
To help detect patterns in these massive datasets, NASA FDL's team relied upon Google Cloud AutoML,
a suite of tools that simplifies the machine learning process for people with limited expertise.
Using Cloud AutoML's resources helped researchers root out false positives, rapidly classify
light curves, and identify key variables they hadn't noticed yet.
"Machine learning can tell us in the blink of an eye what is a planet and what isn't,"
said Hugh Osborn, an astronomer on the exoplanet team.
It didn't take long for researchers to notice the benefits of Cloud Auto ML. Running a single
retrieval used to take researchers several days at a 94% accuracy rate. With Cloud
Auto ML backed by Compute Engine, retrievals ran in seconds at 96% accuracy. That hundreds of
retrievals could be run simultaneously with Cloud Auto ML added even more value for NASA FDL's