Bruce (Zhi) Wen

I am part of the applied research team at Mila, where we solve real-world problems with AI. Example projects include working with Dialogue on automatic symptom detection, with D-BOX on movie event detection, and with AlayaCare on agent evaluation.

These days I work a lot on building and evaluating agents for various use cases. Fittingly, DDXPlus, which we released in 2022 from the Dialogue project, has become a popular benchmark and training dataset for medical LLMs and agents. Personally, I am most interested in a few topics in NLP, such as generalization of self-improving agents, alternatives to current LLM recipes, controllable generation, and decoding algorithms. I also keep an eye on healthcare/bio applications, such as protein design.

I earned my master’s degree at McGill University, and I did my undergrad at Wuhan University, China. My research focused on machine learning and NLP for healthcare. For instance, I worked on COVID-19 media news surveillance for public health measures, constructing a large medical NLP pre-training dataset, among others.

In my free time, I’ve recently gotten into rock climbing. I used to play a lot of football (the one where you play with your foot). I have an eclectic taste in music, but what I listen to mostly clusters around: progressive, post-rock, alternative, and ’60s–’80s Japanese music.