Zijia Li

About Me

I am an MSE student in Data Science at the University of Pennsylvania. I am also a master’s student with the Generative Machine Learning Research (GMLR) Lab at Penn’s GRASP Laboratory, where I work with Prof. Jiatao Gu. My current research focuses on generative world models, particularly how learned latent representations and generative dynamics can support planning and decision-making in embodied agents.

I am currently applying for Ph.D. programs starting in Fall 2027. I hope to pursue research at the intersection of world models, embodied AI, and generative modeling.

Research Interests

• World Models & Embodied AI: representation learning, latent dynamics, and model-based planning
• Generative Models: flow matching, diffusion models, and generative learning
• Foundations of Machine Learning: mathematical perspectives on representation, uncertainty, and generative modeling

Education

• University of Pennsylvania, MSE in Data Science, 2025-2027 (expected)
• University of Nottingham / University of Nottingham Ningbo China, B.S. in Mathematics with Applied Mathematics, 2021-2025

Research Experience

• Master’s Student, GMLR, Penn GRASP Laboratory
  Developing generative world models with flow-based latent representations. My work explores generative extensions of JEPA-style models, multimodal future prediction, and planning under uncertainty through latent rollouts and learned value functions.
• Research Assistant, Intelligent Computing and System Lab, Zhejiang University
  Investigated flow matching and parallel sampling methods for diffusion models, with an emphasis on probability paths and efficient generation.
• Research Assistant / Intern, University of Nottingham and Northwestern University
  Worked on PINNs for wave equations, diffusion-model fine-tuning, and Transformer-based in-context learning.

Selected Projects

• Image2GPS: Built a campus camera-location prediction model using ConvNeXt, grid-based classification, and coordinate regression.
• Multimodal Stock Price Prediction: Developed a forecasting pipeline that integrates market data, fundamentals, and sentiment features.
• River Flow Prediction: Designed a hybrid SARIMA-LSTM model for daily river-flow forecasting in the UK.

Selected Honors

• Dean’s Scholarship, University of Nottingham Ningbo China
• Zhejiang Provincial Scholarship
• Honorable Mention, Mathematical Contest in Modeling
• Third Prize, Formula Student Electric China 2022