Is federated learning a privacy breach, given that the model transmission, for a period of time, may cause the adversarial to reach and manipulate the model and the data?

  • 1
    $\begingroup$ According to the literature, there are many privacy concerns available. The literature about how to backdoor distributed machine learning, how to secure the communication and create private recurrent neural networks is large. The first papers were published since the year 2017, so it's a relative new research field. $\endgroup$ – Manuel Rodriguez Sep 20 '19 at 10:54
  • 1
    $\begingroup$ Are you asking whether federated learning is a privacy breach or not? I think so, thus I edited your question to reflect this. $\endgroup$ – nbro Sep 20 '19 at 12:20
  • $\begingroup$ @ManuelRodriguez oh, i don't know why someone changed my question, the question is like this : Isn't this a privacy breach of Federated learning? model transmission for a period of time which may cause the adversarial to reach and manipulate the model and kind of data ? $\endgroup$ – Najib jamshidi Sep 23 '19 at 5:07
  • $\begingroup$ @nbro no man, my question is so clear, I am asking that does adversarial can access the model transmission? it mean, while the primary model trained by device will be sent to FL Server and then the server will send back a global model, this will be a round and it may happen several or thousands of time so the attackers can access the model, is it possible that attackers access the data using the model they access? $\endgroup$ – Najib jamshidi Sep 23 '19 at 5:11
  • $\begingroup$ @Najibjamshidi The question was so clear to you, not to me (and maybe others). Now, after your clarification, it seems clearer. Just edit your post to add these explanations! $\endgroup$ – nbro Sep 23 '19 at 8:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.