Senior Research Scientist — Self-supervised Representation Learning
Donghyun Park
Current focus: Pretrained driving-video backbone that beats supervised init on nuScenes detection with 10× less labels.
Background
PhD NYU 2022, advised by Yann LeCun and Carlos Fernandez-Granda; thesis on joint-embedding predictive architectures for video. Research Scientist at Meta FAIR (2022–2025) contributing to DINOv2 and the V-JEPA driving-video pretraining recipe. Lifelong Bay Area cyclist who pretrains on his own GoPro footage on weekends.
Education
- PhD · Data Science / Computer Science — New York University (2022), advisor: Yann LeCun
- BS · Statistics & CS — Yonsei University (2015)
Selected publications
- Park et al., 'Driving-JEPA: Joint-Embedding Predictive Pretraining for Driving Video', NeurIPS 2023
- Park et al., 'Scaling Self-Supervised Pretraining to 100M Hours of Driving Video', ICLR 2024
- Park & LeCun, 'Spatiotemporal Predictive Features Transfer to Closed-loop Driving', NeurIPS 2024
Reading list
- He et al., 'Momentum Contrast for Unsupervised Visual Representation Learning (MoCo)', CVPR 2020
- Caron et al., 'Emerging Properties in Self-Supervised Vision Transformers (DINO)', ICCV 2021
- He et al., 'Masked Autoencoders Are Scalable Vision Learners (MAE)', CVPR 2022
- Oquab et al., 'DINOv2: Learning Robust Visual Features without Supervision', TMLR 2024
- Bardes et al., 'V-JEPA: Latent Video Prediction for Visual Representation Learning', ICLR 2024
Lead experiments
All experiments →No experiments led by this member yet.