REVIEWER 1:$ 
1-  Pinto’12 was the first work to study the problem of locating the source from a random set of observer nodes that can collect time stamps. Indeed, we use the algorithm developed in Pinto’12 to locate the source under diffusion. The work in Farajtabar et al., AISTATS 2015 is also relevant, and we would be happy to cite it.
2-  The spreading algorithm used in Section 2 of our paper is a special case of the adaptive diffusion algorithm presented in Fanti’15 (Section 3). It’s different from the spreading algorithm presented in Section 2 (Spreading on Line) of Fanti’15. We will clarify this point in the final version.
3- In our paper, we don’t use the Gaussian assumption to diffuse information on a social network. We use the original diffusion for that. The Gaussian assumption was used in Pinto’12 to derive a source estimator (which we refer to as the Gaussian estimator). We use the Gaussian estimator to locate the source under regular diffusion. We will clarify this in the final version.
 
REVIEWER 2:
 
The novelty of this work is not as much in the spreading protocol which was proposed earlier although in a significantly different adversarial setting. However, it is not at all clear whether this protocol would be effective against spies with timing information, and what parameters one should choose to get the best performance. To our surprise, we identified a choice of the parameter that achieves (asymptotically) optimal source obfuscation. Further, the theoretical analysis presented in this work /is/ novel and gives the surprising result: adaptive diffusion, which was designed for a very different adversarial model, is asymptotically optimal and hides better as the node degree increases, whereas the opposite is true of regular diffusion. 
 
Thank you for your comments. We will move Section 2 to the supplementary material and incorporate your comments in the final version.
 
REVIEWER 3:
 
Regarding interest to NIPS community: We pose a question of making inference sufficiently hard for the purpose of protecting privacy. This theme might be of broader interest to the NIPS community, and possibly present new directions in other inference problems than the specific source detection problem addressed in this paper.