Sandia National Labs Academic Programs Collaboration Report

Modern ML techniques have proven very effective at finding objects of interest within a scene, but when it comes to rarely seen objects, it becomes significantly more challenging since ML techniques employ algorithms that require a large number of representative images for training. Sandia and Prairie View A&M University collaborated on a project to better understand how limited training sets can be leveraged for such algorithms via a process known as data augmentation. Through data augmentation, researchers increase the amount of data by using existing data and then creating slightly modified copies with minor alterations, or synthetic data. This synthetic data expands the amount of information available for training. This research team discovered that while data augmentation does have limitations, it still shows promise in improving the accuracy of the ML detection algorithm. Utilizing data augmentation to locate rarely seen objects

103

2021-2022 Collaboration Report