Weakly-supervised Autonomous Learning for Detecting Objects (WALDO)

Jaxon, Inc.’s commercial platform JAXON with AI driven Natural Language Processing (NLP) and Machine Learning (ML) capabilities for automating sparse-label training datasets will be expanded to provide the Air Force with Weakly-supervised Autonomous Learning for Detecting Objects (WALDO) that will support deep learning image segmentation and object detection. As a semi-supervised platform, hybrid labeling approaches will seek to optimize the combination of a small number of detailed expert annotations, a larger number of fast non-expert annotations, and a pipeline of self-supervised learning techniques that can be bootstrapped from these initial annotations. Technologies such as active learning, curriculum learning, dataset similarity measurement, and weak supervision will be strategically combined to maximize limited human input to learn subtle image characteristics. This effort directly supports the U.S. Air Force 2030 Science and Technology Strategy for Rapid, Effective Decision-Making by increasing autonomy in tactical-to-strategic environments and overcoming unpredictability and uncertainty in combat.