Finding Label and Model Errors in Perception Data With Learned Observation Assertions

A neural network system that is trained on the general properties of objects (ex. size and speed) and then analyzes a dataset to attempt to find errors that previous classification efforts left in the training data.