[UG/MS/PhD] Student Research Opportunity — Generative AI for Forecasting Real-World Time-Series & Dynamics

The following announcement is from [Hao Cheng]. Please contact them directly if you have any questions.

We are recruiting motivated undergraduate and graduate students to join a research project at the intersection of modern generative AI and time-evolving real-world data. The goal is to develop models that learn dynamics from data and generate realistic future rollouts—especially in complex systems where uncertainty is intrinsic.

No physics background is required. If you are comfortable coding and curious about machine learning, you can ramp up quickly.

Project Motivation

Many real datasets evolve over time in complicated, noisy, and partially observed ways. Standard forecasting models often drift over long horizons or produce overly smooth predictions. In this project, we aim to build generative models that can:

• Produce multiple plausible futures (uncertainty-aware forecasting)
• Reconstruct incomplete, corrupted, or low-resolution data
• Generate realistic samples for downstream analysis and scientific discovery

What This Enables

Potential outcomes include methods that support:

• Forecasting future evolution with uncertainty quantification
• Super-resolution from coarse measurements
• Denoising and reconstruction for missing/corrupted observations
• Realistic simulation and synthetic data generation for analysis

Example Application Areas (Choose What Interests You)

Students may work in one or more of the following domains:

• Weather and flow-related spatiotemporal forecasting (structured space–time fields)
• Medical or biological imaging time series (reconstruction, denoising, forecasting)
• Astronomy and other scientific datasets with temporal structure
• Any high-dimensional time-evolving dataset with uncertainty and complex dynamics

What You May Work On

Depending on your background and interests, responsibilities may include:

• Implementing and evaluating generative forecasting models (e.g., diffusion-based approaches)
• Building clean, scalable training/evaluation pipelines (GPU-based)
• Benchmarking against strong baselines and ablations
• Writing short, clear experiment summaries (results, takeaways, next steps)
• Maintaining reproducible, well-documented code and workflows

Mentorship Team

Students will receive guidance on literature review, experiment design, evaluation methodology, and writing.

• Haotian Hang — PhD (USC), incoming Postdoctoral Researcher at IRPHE (Institute for Research into Non-Equilibrium Phenomena). Research: AI for physics, multi-agent systems, reinforcement learning, biological locomotion, unsteady fluid dynamics. Email: hanghaotian@gmail.com
• Jingyi Liu — Theory Fellow Researcher, Janelia Research Campus (HHMI); PhD (USC). Research: microscale and biological fluid dynamics, cell migration/aggregation, decision-making in self-organized systems. Email: jingyiliu1900@gmail.com
• Morgan Jones — Postdoctoral Researcher, Los Alamos National Laboratory. Research: generative diffusion models for emulating hydrodynamic shocks. Email: morganrj@gmail.com
• Hao Cheng — PhD student, Mechanical Engineering (USC).Research: statistical physics, unsteady fluid dynamics, collective behavior. Email: hcheng85@gmail.com
• (Hao Ji) —

What You Gain

• Hands-on research experience in scientific generative AI (including diffusion models)
• A clear project structure with milestones, regular feedback, and research best practices
• Co-authorship opportunities (including potential first-author papers), depending on contribution

Logistics

• Open to UG / MS / PhD students
• Expected commitment: ≥ 15 hours/week
• Access to computational resources via USC CARC (GPU/HPC)

How to Apply

Please email the following:

1. Short introduction (3–6 sentences) describing your background and interests
2. Resume/CV
3. (Optional) GitHub and/or links to prior projects

We will follow up to schedule a brief chat. If there is a good fit, we will help you get started quickly with an initial reading list, baseline experiments, and a first milestone.

 

Published on March 31st, 2026Last updated on March 31st, 2026