1
Paid research for PhD and Postdocs this summer
“I think FDL is uniquely positioned to put the right people on the job at the right time and producing ‘firsts’ in terms of applications.
2
Atilim Güneş Baydin
3
AI for lunar exploration
Over the past 7 years, NASA, The Luxembourg Space Agency / Space Resources and FDL.AI have had a hugely impactful research partnership leveraging AI for lunar exploration and resources, with notable firsts in advance of the Artemis mission.
The formula of data science specialists working closely with subject matter researchers has been remarkably consistent, with many peer reviewed research outcomes and data products, from rover localization without GPS to co-operative robotics and crater-counting for impact age determination. Some of our most impactful work for Artemis is shared below.��Notable firsts:�Lunarlab’s research used AI to peer into the permanently shadowed regions (PSRs) of the Moon to give us a clear view into the permanent darkness for the very first time. HORUS continues to be used for traverse planning.
Lunarlab’s work on thermal anomaly detection revealed the location of the Moon’s metallic resources using a breakthrough application of Physics Informed Neural Nets called a PINN. This AI allows us to see the Moon’s resources clearly.
Lunar South Pole Permanently Shadowed Regions (PSRs)
4
5
“I would have to say it's the model of rapid development plus the guts to tackle super big problems. It's quite an adaptive model. I have not seen it elsewhere and I tried to propose it to some universities or to the National Academies of Sciences. In academic science, things tend to more much slower and researchers are more likely to take on incremental problems instead of large ones, because they also need to maximize their chances of publishing.
”-Anamaria Berea
6
“What really made FDL special for me was the community, collaborative atmosphere the group was, with someone always willing to help and the fact your working in cutting edge problems”-Laura Hayes
7
“Making the right connections in the right setting. Connecting people in general. Connecting machine learning researchers with space scientists, connecting these researchers with industry who provide resources, then connecting all of these with space agencies, etc.”
-Atilim Güneş Baydin
8
“What I've been telling people is that the model of putting scientists and ML experts at the same desk for 2 months to work on a specific problem is the best way (in my opinion/experience) for scientists to really understand how to use ML properly in their research. Too often you see scientists trying to apply ML where it shouldn't be applied (e.g., where there isn't a good human-labeled dataset, so they just use models for training) , or not taking care to avoid common pitfalls in ML research (e.g., not putting aside a test set in the beginning).
There is so much nuance in applying ML to scientific problems and the focused, small-team approach of FDL really facilitated rapid learning in the best way possible, I think. I tried several times to learn ML "on the side" and failed because 1) it's hard/complicated, 2) postdocs and grad students are already over-worked, and 3) it's very different from traditional science. I know of a lot of colleagues who have or are going through the same thing. The intensive approach of FDL really helps to get you over the large learning curve as fast as possible. It was pretty amazing.”
-Megan Ansdell
9
Key Details
10
12
Thank you.