Shobolinsky at ESA: A Hackathon to Remember

The rats recently participated in a Hackathon organized by ESA. The task? Train a machine learning model that detects the composition of exoplanets’ atmospheres when transiting in front of their home star. The challenge lay in the sheer size of the dataset—over 2,000 exoplanets’ worth of data to work through. But we didn’t back down. In fact, we charged full throttle into the task, even when the odds seemed stacked against us.

We trained the machine learning model through the night, powered by nothing more than caffeine and determination. After two days of intense focus, a sleepless night, and plenty of debugging, our model was born. We weren’t about to complicate things with the latest and greatest algorithms. Ridge regression? Nah. We went for something simpler, something humble: linear regression. It might not have been the most sophisticated tool in the box, but sometimes simplicity is key, especially when you're on a time crunch.

Despite our naive approach, we started seeing results. It wasn’t flawless, of course, no first attempt ever is. But it was enough to keep us going. Along the way, we faced some serious hurdles.

Sleepless nights? That was just part of the deal. But the real challenge came from the immense pressure of competing alongside people who had way more experience than we did. We weren’t just the underdogs, we were the youngest, and arguably the least experienced, team in the room, surrounded by scientists, engineers, postdocs, and seasoned researchers. It felt like we were swimming in deep waters without even knowing how to tread.

But that didn’t deter us. We had read the Ariel research in preparation for the Hackathon, and armed with this knowledge, we anticipated that there might be some drift in the data. We made sure our model was prepared to handle that variability, and it turned out that we were the only team to consider this aspect of the dataset. In hindsight, it was a crucial step, and one that set us apart.

One of the key moments of this journey was realizing that we were onto something. Our model, though simple, was beginning to detect patterns in the vast dataset of exoplanet transits. It was showing us how even small, seemingly insignificant variations in the data could make a big difference in understanding the composition of distant worlds.

But let’s be real for a moment, no journey is without its funny moments. One of the funniest was when we decided to experiment with some unusual hyperparameters: 6, 9, -1, 6, and 9. It seemed like a random combination, but to our surprise, it gave us a result that was only 2 points behind the best on the leaderboard. We couldn't help but laugh at the coincidence—it felt like the universe was in on the joke. It was a moment of pure absurdity, but it worked, and that little quirk in our approach showed that sometimes, even the most unconventional solutions can lead to surprisingly good results. When we got it right? It was a moment of validation that made all the chaos feel worth it.

However, just when we thought we had it in the bag, we experienced a moment of devastation. As we watched the public leaderboard, we saw our ranking drop from 2nd to 5th. The pressure hit hard, we thought, “Well, at least we came, saw what this is all about, and gave it our best.” We were ready to go home with no victory. But that wasn’t the end of the story. When the results with the hidden score were revealed, we discovered that our model hadn’t overfitted. In that moment, we were in a state of pure joy and disbelief, realizing that our work had paid off in ways we hadn’t expected. The excitement of seeing our model perform well on the final dataset was the ultimate reward for all the sleepless nights and hard work.

Then came the big surprise: we landed second place. Yes, second! We had gone from having no experience with AI and machine learning to finishing near the top of the leaderboard, all within 30 hours. It wasn’t just about the model we built, it was a testament to our resilience, creativity, and the power of learning on the fly.

But perhaps the most unforgettable part of the experience was simply being there at ESAC. We met some incredibly clever people, made lasting friendships, and found ourselves in awe of the environment we were in. It took us some time to process the whole experience—we still couldn’t believe that we were actually at ESA, the ESA, the European Space Agency, surrounded by the best and brightest minds in space exploration. It was the kind of experience that takes a while to sink in, and one that we’ll remember forever.

What did we learn from this experience? First, that even when you're working with limited resources and no expertise, persistence can take you a long way.

And second, that failure is an important part of the process. Every mistake we made, from incorrect predictions to data mishandling, was a chance to grow and improve.

So here’s the takeaway: If you’re ever in a situation where you feel like an outsider—where everyone around you seems more qualified and experienced, don’t let that intimidate you. Dive in, learn as much as you can, and give it everything you’ve got. You might just surprise yourself with what you can achieve.

And who knows? Maybe next time, we’ll be gunning for that first-place finish.