Skip to content. | Skip to navigation

Personal tools

Navigation

You are here: Home / Meetings / All Meetings / NASA Frontier Development Lab 2018

NASA Frontier Development Lab 2018

FDL is an applied artificial intelligence research accelerator established to maximize new AI technologies and capacities emerging in academia and the private sector and apply them to challenges in the space sciences. It is an interdisciplinary research program at the Phd and Post-Doc level. Our researchers come from around the world to work on unresolved problems in the space sciences identified by senior NASA scientists. The 8-week summer program places an emphasis on learning and rapid iteration, cross-disciplinary skills sharing, prototyping and experimentation. The 2018 Challenges cover space resources, exoplanets, space weather, orbital debris, and Earth observation. The exoplanet challenges are oriented towards the TESS mission. Researchers receive a stipend, accommodation and visa support.
When 25 June 2018 08:35 PM to
17 August 2018 08:35 PM
Where Mountain View CA USA
Add event to calendar vCal
iCal

FDL 2018 EXOPLANET CHALLENGES

Challenge 1: Increase the efficacy and yield of exoplanets detection from TESS

TESS’s mission to discern credible exoplanet signatures is an imposing data task. Moreover any follow-up strategy to further examine interesting candidates is made more demanding because of a mission time constraint, where follow-ups need to be identified each and every month as the spacecraft changes its field of view. At the moment, diagnosis of Exoplanet candidates still requires manual analysis - an impossibly big challenge for humans in the time window available. This time crunch problem remains unresolved, but may be suited to AI approaches. Can we develop AI techniques to resolve the time window problem, by automating the classification analysis currently done by human experts? 

Challenge 2: Codify the process of AI derived exoplanet discovery and classification to help TESS better discern rocky planets. 

Can we use data from Kepler, K2 and and human analysis to develop a method to improve TESS’s ‘Lilith’ simulator factoring a range of stellar properties and occurrence rates and well as develop an ‘explainable AI’ that can learn to better discern rocky classifications as time goes on and identify the key features factored in the analysis? In addition to possible improvements to Lilith this challenge would look to identify improvements to the detection pipeline, not just in the transit detection but also upstream data reduction. I.e. what pixel calibration errors are most detrimental? how best to remove systematics? which systematics cause the most issues? and also identifying improved vetoes in the transiting planet search module.

More information about this event…