PhD student · CLSP · Johns Hopkins
Xiutian (Theo) Zhao
I look inside audio-language models to see how they represent human concepts, like emotion and speech-based diagnostic cues, and turn what we find into something utilizable.
About
I’m a PhD student in Electrical and Computer Engineering at Johns Hopkins University, affiliated with the Center for Language and Speech Processing (CLSP) and the Data Science and AI Institute (DSAI). I’m fortunate to be advised by Dr. Berrak Sisman in the SMILE Lab, and I’m grateful for the support of the Amazon AI PhD Fellowship.
My research focuses on utilizable mechanistic interpretability of audio-language models, particularly how they represent human-centered concepts and knowledge, such as emotion and speech-based diagnostic information. Earlier work explored spoken-language understanding and generation, LLM agent learning, and collective decision-making in multi-agent systems. For my master’s dissertation at the University of Edinburgh, I worked with Dr. Ivan Titov, Dr. Rochelle Choenni, and Dr. Rohit Saxena.
Outside research, I’m a cinephile (2,000+ films and counting) and enjoy landscape photography, swimming, stargazing, and reading ancient history.
Selected papers
All 20 publications →Discovering and Causally Validating Emotion-Sensitive Neurons in Large Audio-Language Models
News
I’m honored to be named a 2026–2027 Amazon AI PhD Fellow through the JHU + Amazon Initiative for Interactive AI (AI2AI)! ✨
1 paper accepted to ACL 2026! See you in San Diego! 🐬
1 paper accepted to EACL 2026! I will volunteer at the conference, see you there!
I graduated from the University of Edinburgh
An adaptation of my master’s dissertation is now on arXiv! Many thanks to Dr. Rochelle Choenni, Dr. Rohit Saxena, and Dr. Ivan Titov.I’ll start my PhD at Johns Hopkins University in Fall 2025! Super excited to join Dr. Berrak Sisman’s group! 🎉