Heejun Yoon

Welcome! 👋

Hi, I’m Heejun Yoon. I’m a Ph.D. student at Mila and HEC Montréal, advised by Tegan Maharaj and Olivier Bahn. I started in August 2026, and my Ph.D. is supported by the Canada Impact+ Research Training Award.

I want to build AI that helps with environmental and conservation problems. Right now I’m thinking about how climate change and biodiversity loss connect, and how a model could help people see the downstream effects of a climate policy before it is put in place. The topic is still taking shape; you can find my early notes on the projects page.

Before Montréal, I was a researcher at the Korea Institute of Science and Technology (KIST). I hold M.S. and B.S. degrees in Electrical Engineering from Ewha Womans University, with a minor in Mathematics.

Research interests

Most modern AI does well on large, clean benchmarks. Ecological data looks nothing like that. It is sparse and expensive to collect, it is biased toward places that are easy to reach, and it is messy. That gap is the problem I care about most.

Two questions keep coming back:

• Fusing fragmented data. Understanding an ecosystem means combining satellite images, acoustic sensors on the ground, drone footage, and sometimes text like policy reports. How do we build multimodal models that combine these imperfect sources into one estimate that people can act on?
• Decisions where errors have real costs. In conservation, a wrong prediction can mean a habitat nobody protected or money spent in the wrong place. How do we build decision frameworks that weigh those costs directly, instead of only reporting an error rate?

I’d like the tools I build to be useful to the scientists and policymakers who make these calls, and honest about what they don’t know.

How I got here

I started with 1D radar signals as an undergrad, detecting human motion and vital signs. My master’s work moved to 2D machine vision, where I built models that find tiny defects on semiconductor substrates. At KIST I worked on 3D scenes and embodied agents: vision-language models that reason over multiple views, and vision-language-action models that have to follow long instructions in simulated homes.

Each step taught me something about noisy, real-world data. Now I want to point that experience at climate and biodiversity.

If you’d like to talk about research or anything else, email me (heejun.yoon@mila.quebec or heejun.yoon@hec.ca) or find me on LinkedIn.