Ting-Wu Chin

Carnegie Mellon University, National Yang Ming Chiao Tung University

Papers

Total Citations

H-Index

About

Ting-Wu Chin is a computer vision and machine learning researcher whose work sits at the critical intersection of model efficiency and real-world deployment, particularly for resource-constrained autonomous systems. His research focuses on two core challenges: accelerating video object detection and developing principled neural network compression techniques. Chin's most recognized contribution, AdaScale (2019, 39 citations), challenged the conventional wisdom that detection speed and accuracy are inherently at odds, introducing adaptive scaling strategies that bring real-time video object detection closer to practical deployment in autonomous vehicles and robotics. Complementing this, his work on domain-specific approximations (2018) further explored accuracy-speed trade-offs tailored to driver assistance systems. On the model compression front, Chin developed LeGR and its extended framework (2019–2020, combining over 40 citations), which introduced learned global ranking for filter pruning in convolutional networks. These methods addressed a significant gap in prior art by eliminating the need for users to manually specify target model complexity, making compression more accessible and automated. Collectively, Chin's contributions advance the practical viability of deep learning in safety-critical, real-time applications, making him a notable voice in efficient computer vision research.