This is a hybrid event with in-person attendance in AGH 306 and virtual attendance via Zoom.
The last few years are witnessing a major shift in the way we think of computing, with humans relying more and more on AI assistants which speak their language and understand images and videos. While this revolution has largely been possible by wrapping the world with a digital layer allowing artificial systems to natively ingest information and produce content, future AI systems will need to conquer the same physical space in which humans live. Towards this direction, egocentric vision has established itself as a powerful paradigm, placing a camera where humans have eyes, making it possible to perceive the world from their unique point of view. In this talk, I will first discuss the role that the community envisions for egocentric vision in the AI revolution, then highlight the foundational role of large-scale egocentric datasets, such as EPIC-KITCHENS, EGO4D, and Ego-Exo4D in advancing our ability to understand human behavior. I will next present research efforts aimed towards developing deep learning models to perceive, understand, and anticipate human actions and interactions from this unique first-person perspective. Finally, I will discuss how these capabilities can pave the way for assistive technologies on wearable devices designed to provide direct support to users in procedural activities.