Research in gathering and reasoning about common sense knowledge has recently gained in popularity. Measuring the progress of research in this multimodal space faces many challenges. In this talk we discuss the lessons learned, and remaining challenges in creating testing datasets and associated evaluation metrics.
Larry Zitnick is a research manager at Facebook AI Research in Menlo Park, and an affiliate associate professor at the University of Washington. Previously, he spent 12 years at Microsoft Research. His current areas of interest include object recognition, language and vision, and methods for gathering commonsense knowledge. He is one of the co-organizers of the very successfulMSCOCO challenge.