Me, AI; You, Human—Advances in Human-AI Cooperation

The seminar is located at 2300 the Ford Robotics Building

Abstract
Powered by advances in algorithms, data resources, and hardware, we are in the midst of an AI revolution making indubitable progress.  However, if we inspect this progress through a lens focused on the role the human plays within these AI systems, we find surprising insights.  Namely, fully-supervised, one-time-human-involvement scenarios are significantly over-emphasized in research whereas human-AI cooperative scenarios have made great practical strides toward real-world application.

This talk will journey through a sampling of my research that challenges the status quo of human-AI cooperation in AI research, specifically emphasizing problems in the visual domain.  First, I will present a method, called ClickHere CNNs, that asks a human for a single piece of information, which is then directly incorporated into an inferential engine for estimating 3D object pose from a single image.  Second, I will introduce a new problem that we call Active Clustering; like active learning, we partner with a human to iteratively ask for guidance to increase our labeled data.  However, unlike active learning, where one explicitly asks for labels, in active clustering we seek grouping constraints.  Although active clustering has a higher complexity, it is a more general problem with broader applications.  I will describe our innovative algorithm for active clustering based on spectral learning and matrix perturbation theory.  Third, I will present our recent work in video object segmentation that automatically selects the best frame in a video to use when asking a human partner for guidance. Our method, called BubbleNets, learns a frame-comparison function that is used in sorting frames for their relative value in video object segmentation.  Time permitting, I will also cover recent analytical work on measuring the quality of the human input to a certain task and whether or not to sequentially re-query the human for more information.

Bio:
Corso is Director of the Stevens Institute for Artificial Intelligence and Brinning Chair Professor of Computer Science at the Stevens Institute of Technology, and Co-Founder / CEO of the computer vision startup Voxel51. He received his PhD and MSE degrees at The Johns Hopkins University in 2005 and 2002, respectively, and the BS Degree with honors from Loyola College In Maryland in 2000, all in Computer Science. Prior to Stevens, he was Professor of Electrical Engineering and Computer Science at the University of Michigan, Associate Professor of Computer Science and Engineering at SUNY Buffalo, and a Post-Doctoral Fellow at the University of California, Los Angeles. He is the recipient of a U Michigan EECS Outstanding Achievement Award 2018, Google Faculty Research Award 2015, the Army Research Office Young Investigator Award 2010, NSF CAREER award 2009, SUNY Buffalo Young Investigator Award 2011, a member of the 2009 DARPA Computer Science Study Group, and a recipient of the Link Foundation Fellowship in Advanced Simulation and Training 2003.  Corso has authored more than 150 peer-reviewed papers and hundreds of thousands of lines of open-source code on topics of his interest including computer vision, robotics, data science, and general computing.  He is a member of the AAAI, ACM, MAA and a senior member of the IEEE.