Tianyi Xu

I’m Tianyi Xu, a senior at the University of Wisconsin–Madison (B.S. in Computer Science, Data Science, and Mathematics; expected Dec 2025). I am advised by Prof. Junjie Hu and Prof. Claudia Solís-Lemus at UW–Madison and intern with the EPIC Lab at Shanghai Jiao Tong University, advised by Prof. Linfeng Zhang. I’ve previously collaborated with Prof. Pedro Morgado and Prof. Chaowei Xiao, and I’m fortunate to be mentored by Evan Gorstein, Jiongxiao Wang, and Shaobo Wang.

I study data-efficient, controllable multimodal learning and invariance-driven representations, with applications to real-world problems such as bioacoustics and biomedicine. Specifically, my work focuses on:

1. Representation Learning — Building invariance-driven multimodal representations across vision, audio, and language.

2. Data-efficient and Data-Centric Learning — Learning from few/weak labels via self-supervision and weak/active supervision; LLM pretraining with data selection and domain-mix adaptation.

3. Controllable Generative Models — Steering latent diffusion/rectified flow with conditioning for precise guidance under tight compute budgets, including vision-guided audio separation.

4. Applications — Tone-robust speech representations (e.g., supporting hearing-impairment diagnostics), bioacoustic detection for biodiversity monitoring, and data toolkits/benchmarks for biomedical and clinical settings.

I’m seeking Ph.D. opportunities for Fall 2026.