I'm starting to do research in machine learning on sensitive datasets, which sometimes seems like a double-edged sword. For instance, some projects involve showing how a company/organization/government can infer sensitive information about you and abuse it. I feel that this has to be published as it may already be abused somewhere or will be soon. At the same time, if I publish too many details then some corporations etc. may find it easy to implement and start abusing.

Do some conferences/journals accept papers that have scarce details on some parts of the modeling such that it may hinder abuse? It should be noted that while the code may be sent to the reviewers for a check, the data is sometimes of a sensitive nature such that it cannot be handed over.

  • 1
    You must read the related literature and not speculate on your own (as you seem to do now). All methods for extracting sensitive data from datasets are well-known. Thus, privacy research aims to propose ways to transform the datasets for providing sufficient protection against those attacks (e.g., k-anonymity). So, read the corresponding papers and follow their lead. – Alexandros Feb 16 '16 at 15:39
  • 3
    I think that I may not have expressed myself clearly enough. I'm not doing what you seem to define as privacy research. My works concerns modeling sensitive datasets using machine learning methods (I've updated my question). Here I would definitely disagree with your statement saying "All methods for extracting sensitive data from datasets are well-known". Almost everyday, there's research published on new methods yielding better results on identifying your personality from facebook likes, mental illnesses from mobile phone data etc. This is the kind of research I'm talking about. – pir Feb 16 '16 at 15:54
  • I think there are a couple of unrelated question being asked. What data need to be given to the reviewers is very different from how functional the code needs to be. – StrongBad Feb 16 '16 at 15:57
  • 2
    I think now your post is better. Your focus is on machine learning and data mining. Every paper must always provide enough details for anyone (in this line of research) to be able to replicate your results. If you do not want your method to be abused, do not publish it (that would not stop anyone else publish a similar method though). You cannot get a paper published that proposes a new method for "breaching" privacy and hide the necessary detail to block other people from using it. This is not how science and papers work. – Alexandros Feb 16 '16 at 16:02
  • while your intentions are noble, your desire to restrict the flow of information to prevent potential abuse will likely also hinder your own professional interests. essentially it will prevent other researchers from expanding on your work in a fruitful way. one possible compromise may be to offer scant details and in the methods section, include a line such as "full dataset/model available from corresponding author upon request." this will allow you to disseminate to your peers in a meaningful way. – faustus Feb 17 '16 at 18:19

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Browse other questions tagged or ask your own question.