Sustainability Minor Series: Mara Iris Dumitru

What inspired you to pursue a Sustainability Minor at Minerva? How do you see it complementing your Computational Sciences major? 

I’ve always been a math person. Numbers just make sense to me. But early in my CS degree, especially when we got into Machine Learning, I hit a bit of a wall. I realized I didn't want to just build another app or optimize ad revenue. I needed the math to actually do something. I added the Sustainability Minor to pull my head out of the code and look at the physical world. Combining these fields lets me use data for things that matter. What matters to me is figuring out how to reduce ocean microplastics or cool down urban heat islands.

Why is it essential for the next generation of leaders to understand the long-term impact of sustainable decision-making? 

It’s the irony of the tech industry right now: we have infinite ambition, but finite resources. You see it clearly with AI. We’re training these massive models to solve efficiency problems, but the data centers running them are guzzling water and energy at an unsustainable rate. If we don’t manage the hardware side now, we're going to hit a ceiling that stops progress entirely. Future leaders need to realize that sustainability is beyond just ‘saving nature,’ but also about preserving the resources we need to keep building things.

This summer, you interned at the Tokyo Sustainability AI Lab. Can you tell us about your work with the Blue Ocean Initiative? 

I was tracking plastic waste near Tsushima Island, which is this pollution bottleneck between Japan and Korea. The geography is so jagged that boats can’t really get in to clean it up. In fact, they only manage to recover about 30% of the trash by hand. My goal was to catch the plastic before it hit the coast. I used satellite data (CYGNSS) and LIDAR to build a machine learning model called ConvLSTM. Standard deep learning struggles here because ocean trash isn't static. ConvLSTM handles the spatial side (where the plastic is) and the timeline (how it moves). I added a spatiotemporal attention mechanism so the model could figure out which specific times and locations were statistically significant. We got the error rate down to just 0.0025 and could forecast plastic movement 8 days out. It was a huge win to see the code actually predicting real-world flows.

How did your Minerva coursework help you navigate the challenges of that internship? 

Honestly, coding was only half the job. A design course I took last semester forced me to solve problems for actual people, not just for the compiler. When I joined the Blue Ocean team I had to look at the dataset while considering the physical constraints of the island and the people trying to clean it. I’m still working on linking my AI model with physical simulations to get a better grip on how the plastic moves in the real world.

‍Looking ahead, what area of the tech-sustainability intersection are you most passionate about? 

I'm obsessed with "Green AI." Using tech to fix environmental problems is great, but we need to fix the footprint of the tech itself. Data centers have massive energy and water costs. I want to work on that friction point. If we can make our most powerful tools sustainable, we can solve almost anything.

If you are looking for a university that empowers you to turn computational skills into global impact, start your Minerva application today.

‍