Humans are much smarter than robots today, with the ability to adapt quickly to a new environment and learn a new skill from a few examples, among other amazing things. However, human intelligence comes with many limitations. Human perception is limited by the physical constraints of our sensors. Human’s prior knowledge is limited by the amount of data we are exposed to in our lifetime. Humans are limited by our physical capabilities, such as muscle strength and reaction time.
Ruoshi Liu,
Columbia University
In this talk, we will explore key directions where intelligent systems can achieve superhuman performance and reflect on the lessons learned from these pursuits.
Ruoshi Liu is a Ph.D. student at Columbia University, advised by Carl Vondrick and working closely with Shuran Song and Shree Nayar. His research lies at the intersection of computer vision, robotics, and computer graphics. He has published papers at top conferences, including CVPR, ICCV, ECCV, CoRL, NeurIPS, ICLR, etc. His research received news coverage from New Scientists, Hacker News (#1 trending), etc.. Technology derived from his original work has led to multiple startups in the field of 3D generation, and the open-source models and datasets he developed have been downloaded and used more than a million times by other researchers and engineers in the field.