Does the mixed signals distribution affect the ICA performance? I mean the ability of ICA to get the sources.

  • Assume that the mixed signals follows a Gaussian distribution and of course the sources do not follow Gaussian distributions. Then, this implies that we have large independent components or sources, so will the ICA be able to get all of that components?

  • On the other hand, if the mixed signals follows a Poisson distribution, should that mean something regarding the ICA method? I mean should we expect that the ICA will not be able to extract the sources of that signals in an efficient way?

  • $\begingroup$ "Does the mixed signals distribution affect on the ICA performance?" yes, but I guess you've been expecting that... $\endgroup$ Mar 11, 2017 at 21:28
  • $\begingroup$ Other than that, those are four "?" in one question... I've got the impression you've not read too much about the class of methods ICA is so far, and my gut feeling is that you'd much rather like a good textbook recommendation than you're expecting us to answer your very open-ended questions. $\endgroup$ Mar 11, 2017 at 21:29
  • $\begingroup$ @MarcusMüller then would you recommend a good textbook ?(I've just added the 5th question mark) $\endgroup$
    – hbak
    Mar 11, 2017 at 21:54
  • $\begingroup$ Sounds like a job for Karhunen / Hyvärinen's "independent component analysis", Wiley. Good thing: you can read the intro chapter online and figure out whether it's good for you). Maybe Pierre's 1994 paper already answers a lot of questions, since it nice addresses the effects of statistical dependence and algorithmic realization. $\endgroup$ Mar 12, 2017 at 8:58


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.