We thank all three reviewers for their very thoughtful and detailed comments, with which we largely agree. We will include a detailed example in any revised version. The critique of untested influences refers to potential "unknown unknowns" in the data-27 This is generally not possible in a synthetic system. This is 2599 for networking, 473 for jdk, and 5000 for postgres. We acknowledge there may be issues with low sample sizes and will discuss this in any revised version.

Hessian in the proposed estimator is essential. We will add this result in the future version. The stored parameter is first loaded to the model, and then the gradient for each instance is computed. The comparison is therefore essential to show that our estimator is more suitable for DNNs. We are grateful for a suggestion for clarifying our contribution.

We would like to thank our reviewers for their constructive comments. R1: Not first to look at full meshes. Accordingly, we will limit our claim to deep learning approaches for human pose reconstruction. We will release those, along with pre-trained models. Why does fig 4 show stick figures?

We agree with the reviewer.

C2: Rademacher complexity in this paper refers to its empirical version and we will clarify this in the future version. We use policy evaluation errors to evaluate the quality of model learning. The results in Section 6.2 indicate that The shading on plots refers to the standard deviation over 3 random seeds. We will clarify this in the future version. The word "generation" means that the policy could perform For your choice of "No" to the reproducibility evaluation, we would like to point out that the proof, source We thank you for your insightful suggestion.

We thank all the reviewers. We hope the reviewers could increase the rating if the response addressed your concerns. Theoretically, in Eq. (11), if we take Accordingly, we have revised the relevant sections of the paper by adding pertinent technical details. General Comments: All the learned models that we report performance about are certified monotonic. We will make a note of this in the paper; see also G#1 and G#2.

Firstly, we thank the reviewers for their valuable comments. Whilst it is not reasonable in practice to assume that data is sampled i.i.d. As previously stated, we believe our work forms a first step in achieving this goal. We believe that our theoretical model captures this dynamic. An insurance company may gather information from a customer to better evaluate potential risk.

We thank the reviewers for constructive feedback. "Importantly, the number of components is actually decided by the quality of the augmentation operation: an ideal The accuracies of different algorithms are shown in () . Effectively, such a graph has more edges and better connectivity. We will include this detailed explanation in the future version. Please see Figure 1 for an illustration.