I'm a research scientist at NVIDIA in the Spatial Intelligence Lab with Sanja Fidler, working on generative physical AI. I did my PhD in machine learning at the University of Toronto, advised by David Duvenaud. My research centers on ultra-scalable nested optimization - building the tools that power modern AI models, and using them for things like multimodal generation, hyperparameter tuning, and learning in games. Before NVIDIA, I worked on production AutoML at Google and on multi-agent learning at Facebook (Meta) AI Research with Jakob Foerster.
jonlorraine.com · Google Scholar · LinkedIn · @jonLorraine9
- OmniDreams - a real-time generative world model for closed-loop autonomous vehicle simulation
- MOTIVE (Oral, ICML 2026) - motion attribution for video generation
- LLaMA-Mesh - unifying 3D mesh generation with language models
- LATTE3D (ECCV 2024) / ATT3D (ICCV 2023) - fast, amortized text-to-3D
- Optimizing Millions of Hyperparameters by Implicit Differentiation (AISTATS 2020)
- Self-Tuning Networks (ICLR 2019) - bilevel optimization of hyperparameters
Generative models · bilevel and hyperparameter optimization · meta-learning · multi-agent learning · efficient and multi-fidelity optimization