Featured AI2er: Favyen Bastani

Favyen Bastani is a Research Scientist with PRIOR

Favyen Bastani is a Research Scientist with PRIOR.

What put you on the path to your current role?

During my PhD at MIT, I spent several years working on improved methods for detecting roads in satellite images. We came up with several interesting methods, but early on my focus was all on chasing higher accuracy numbers, with a greater emphasis on the novelty of a method than its practicality. At some point, I took a step back to look at the big picture, and realized that detecting roads was only one component, and I needed to figure out how to use that information to actually update existing maps for it to be useful. Since then, I’ve always kept an eye on the gaps preventing techniques from being deployed in practice, and I’ve accepted that many problems don’t actually require novel solutions. All this experience has led me to my current work on geospatial applications of computer vision at AI2!

Favyen at Multnomah Falls in the Columbia River Gorge.

What’s the most exciting thing that has happened with your work at AI2 recently?

We’ve just announced Satlas, a project that’s been in the works for over a year. Satlas consists of global geospatial data, like the positions of solar farms and offshore platforms, that we’ve computed from satellite images using computer vision. The ReViz team at AI2 built an amazing website for Satlas that includes a map where you can visualize all of the data we’ve released, so check it out if you haven’t already!

How does working at a non-profit differ from your previous work?

The best thing for me has been that the focus is on maximizing positive impact rather than maximizing profit. An excellent example of this is the Skylight team at AI2, who help tackle illegal, unreported, and unregulated fishing by developing a platform for surfacing insights about suspicious vessel behavior. This platform is provided free-of-charge to partners, which include many under-resourced coastal countries. I’ve been very lucky to have had the chance to work extensively with Skylight, and a machine learning model for detecting vessels in satellite images that I helped build was deployed in production last year.

“Queen Anne has hundreds of staircases. I toured 108 of them on a walk one Saturday, but towards the end I realized I’m more of a bridge person!”

What are you looking forward to with your work in the coming months?

There’s so much I want to add to Satlas, it’s a bit overwhelming! First, we’re working on models for more geospatial data products, including maps of crop types and urban land use. Second, we’ve started exploring whether we can generate accurate high-resolution satellite images by combining information from many aligned low-resolution satellite images. If this works, it’ll be incredibly useful, since actual high-resolution images are captured at most once a year at a given location, while we’ll be able to generate high-resolution images every couple of months. Lastly, I’m hoping to work on tools that enable other organizations to more easily train and apply models on satellite imagery, since there are so many niche applications that we’re never going to have time to work on directly.

“When I saw a blog post about this well-designed bike runnel at York Rd S and S Hinds St, I knew I had to visit!”

What is your favorite thing about working at AI2?

AI2 includes both research-focused teams like PRIOR and problem-oriented teams like Skylight (tackling IUU fishing) and EarthRanger (protecting wildlife). This makes it easy for researchers at AI2 to switch between working on fundamental AI challenges and exploring the applications of their work to important problems. I also like the open and collaborative culture at AI2, where we aim to release all of the code, models, and data that we develop.

“Hiking Little Si with my parents.”

What do you consider the most underrated activity or place in Seattle?

I love tall bridges, so despite the car noise, I enjoy walking across the Aurora bridge, Jose Rizal bridge, and Magnolia bridge. Each one offers great views of different parts of the city. The Sounder train is definitely underrated too (many people don’t know it exists); it offers a pleasant trip from Seattle to Tacoma or Everett, although the trains only run during rush hour (hopefully they will expand this in the future, but it may be unlikely with the light rail extensions, which I am very much looking forward to).

A view of Seattle from the Jose Rizal bridge.

Learn more about the PRIOR team at AI2 and the Satlas project.

Check out the open roles available at AI2.

Follow @allen_ai on Twitter/X and subscribe to the AI2 Newsletter to stay current on news and research coming out of AI2.