Jonathan Lorraine is a research scientist at NVIDIA, in the Learning and Perception Research group. His current work is on AI for science: agents that propose, run and verify their own experiments, and the measurement discipline that makes their results trustworthy.
His research spans generative models for 3D, video and audio, amortized and bilevel optimization, hyperparameter optimization, and training-data attribution. He is an author of ATT3D and LATTE3D on amortized text-to-3D synthesis, LLaMA-Mesh on 3D mesh generation with language models, Graph Metanetworks, and Score Distillation Sampling for Audio.
He completed his PhD at the University of Toronto with David Duvenaud, on scalable nested optimization for deep learning.
source & further reading
research.nvidia.com — original article
FusionRelight: Relighting Portraits in Real Time via Hybrid Domain Knowledge Fusion
HorizonRelight: Relighting Long-horizon Videos Consistently via Diffusion Transformers
Scalable Nested Optimization for Deep Learning