Xiaosheng Zhao

About Me

My research focuses on developing and applying machine learning methods for astrophysics and cosmology, spanning both Galactic archaeology and early-universe inference. At Johns Hopkins University, I currently work on spectral foundation models—such as SpecCLIP and its fine-tuned extensions—for cross-survey generalization across LAMOST, Gaia XP, DESI, Prime Focus Spectrograph (PFS), etc. I am also interested in the inference of dark matter profile from dwarf spheroidal galaxies with PFS. My previous doctoral research at Tsinghua University explored the epoch of reionization using simulation-based inference (SBI) and optimal summary statistics, and introduced generative approaches (e.g., diffusion models) for astrophysical image synthesis. Broadly, I am interested in connecting data-driven representation learning with physical interpretability to extract scientific insights from large, heterogeneous astronomical datasets.