Emily Liu

I am a machine learning researcher and engineer interested in developing a scientific understanding of how large models learn, represent information, and generalize. My background spans efficient training and inference, representation learning, and robustness under distribution shift.

More recently, my interests have shifted toward understanding why large-scale models behave the way they do, not just optimizing their outputs. This has drawn me toward mechanistic interpretability and empirical evaluation: probing the internal representations and failure modes of models rather than treating them as black boxes. I see this as a natural extension of production ML work, where understanding model behavior under distribution shift, noisy inputs, and changing data is a practical necessity as much as a research question.

I completed my Master of Engineering in Computer Science at MIT, advised by Dr. Caroline Uhler, where my thesis examined causal representation learning for predicting the effects of genetic perturbations in single cells. I also hold bachelor’s degrees in Computer Science and Mathematics from MIT. Currently, I work at ByteDance on large-scale recommendation and multimodal LLM-based ranking models, focusing on model optimization, debiasing, and production reliability. My broader goal is to connect rigorous scientific understanding with practical, trustworthy, and sustainable model deployment at scale.

Email me for an up-to-date version of my resume :)