Jake Levinson
Papers
1
Total Citations
32
H-Index
1
About
Jake Levinson is a researcher whose work sits at the intersection of deep learning, geometry, and computer vision. His most-cited paper, "An Analysis of SVD for Deep Rotation Estimation" (2020, 32 citations), provides a rigorous theoretical and empirical examination of how singular value decomposition (SVD) and related orthogonalization techniques can be integrated into neural networks for stable, differentiable rotation estimation. This work addresses a fundamental challenge in 3D vision—ensuring that network outputs respect the geometric constraints of the rotation group $SO(n)$—and has become a key reference for practitioners working on pose estimation, structure from motion, and 3D alignment problems. By clarifying when and why SVD-based projections succeed or fail in deep learning pipelines, Levinson has helped bridge the gap between classical geometric optimization and modern end-to-end learning. His contributions are particularly valuable for students and researchers seeking principled ways to incorporate geometric priors into neural architectures, offering both theoretical insight and practical guidance for building more reliable 3D perception systems.
Research Focus
Key Achievements
Top Papers
- 1An Analysis of SVD for Deep Rotation Estimation32 citations · 2020