MIT and IBM Train AI to Understand Dynamic Video Events
The MIT-IBM Watson AI Lab develops new methods to teach computers how to recognize and understand actions unfolding over time in videos.
Researchers at the MIT-IBM Watson AI Lab tackle a major challenge in artificial intelligence by teaching computers to recognize dynamic events in videos. While humans easily identify similar actions like a door opening or a flower blooming, current computer models struggle to understand these visual concepts as they unfold over space and time.
The lab brings together scientists from both institutions to bridge this gap in machine perception. Principal investigators Aude Oliva and Dan Gutfreund lead the development of the Moments in Time Dataset to provide the necessary foundation for training these advanced AI algorithms.
This project represents a broader effort by the collaboration to push the boundaries of artificial intelligence. By focusing on how information processes dynamically, the team aims to build AI systems that interpret the physical world with a deeper, more human-like understanding.