Abstract
News editors obtained the following quote from the background information suppli ed by the inventors:“Collecting real-world data for training and testing machin e learning models (MLMs)-such as neuralnetworks-is a laborious, costly, and tim e consuming task that requires countless human and computeresources. Even where sufficient resources are available, certain scenarios that should be captured t oproduce a robust model may be rare or unsafe to capture. To combat this issue, synthetic data generationhas emerged as a solution to generate ground truth in formation. Synthetic data may be generated usingthree-dimensional (3D) graphics techniques to simulate the real-world. In some approaches, real-world datamay be transformed to generate the synthetic data for training or testing, such as t o augment trainingsets. However, synthetic data can include artifacts that are correlated with the ground truth informationso that an MLM may make predictions based on the artifacts. When deployed in the real-world, theseartifacts may no t be present, resulting in a poorly performing MLM.
NETL
NSTL