My talk is going to focus mostly on the third chapter of Ian's thesis. If you read the first linked paper that should give you a good background to understand the third chapter (which extends the learning of objects from the supervised to the weakly-supervised setting).