Based on the information provided, I would encourage the following changes be made to the paper in addition to what was suggested in the original reviews: * Improve comparison with related work by Suciu et al. and Koh and Liang (using points on the more realistic treat model and also experimental results mentioned in your feedback). They also don't require the presence of a trigger at test time (as done in what is commonly referred to as a "backdoor"). To the best of my knowledge, it includes pointers to relevant prior work both on shallow and deep learning algorithms. For these reasons, I think the manuscript is original and should contribute to increased visibility of poisoning in the adversarial ML community. Furthermore, the manuscript is well written and easy to follow.

Data poisoning is an attack on machine learning models wherein the attacker adds examples to the training set to manipulate the behavior of the model at test time. This paper explores poisoning attacks on neural nets. The proposed attacks use clean-labels''; they don't require the attacker to have any control over the labeling of training data. They are also targeted; they control the behavior of the classifier on a specific test instance without degrading overall classifier performance. For example, an attacker could add a seemingly innocuous image (that is properly labeled) to a training set for a face recognition engine, and control the identity of a chosen person at test time.