Tom Goldstein

Papers

Total Citations

H-Index

About

Tom Goldstein is a leading researcher in machine learning, with a focus on optimization, adversarial robustness, and 3D perception for autonomous systems. His foundational work on "Democratic Representations" (2014, 27 citations) introduced a novel approach to minimizing the ℓ∞ norm under linear constraints, advancing applications in vector quantization and nearest neighbor search. More recently, Goldstein has tackled the critical challenge of robustness in autonomous systems; his 2021 paper on "Adversarial Differentiable Data Augmentation" (16 citations) develops methods to fortify neural networks against degraded input images, a key vulnerability for robotic systems. In 2024, his work on "Hierarchical Point Attention for Indoor 3D Object Detection" (6 citations) pushes the boundaries of transformer architectures for point cloud processing, enabling more accurate detection for augmented reality and domestic robots. Goldstein's research bridges theoretical optimization and practical deployment, with his contributions to adversarial machine learning and 3D vision directly impacting the safety and reliability of autonomous technologies. His work continues to shape how neural networks perceive and interact with the physical world.