Weakly-supervised learning from ambient traffic data
Distributed acoustic sensing is a technology that captures real-world acoustic and vibration events at the speed of light. Just as a lighthouse serves as a reference point for navigating ships across the vast ocean, manholes can be employed as landmarks to determine the locations of other detected acoustic events around the cable.
During the summer of 2021, I had the privilege of supervising an internship project focused on localizing manholes within existing networks through learning from ambient data. Our team developed an innovative solution based on deep multiple instance learning, pushing the boundaries of utilizing ambient environment data for fiber sensing tasks with less human annotation. In addition, manhole also serves as the key accessing point and this approach holds great promise in enhancing operational efficiency of telecom carriers and reducing field work.