Multinomial logistic regression have assume the IIA (independence of irrelevant alternatives (IIA), IIA assumption: difference logit and probit); but I'm wondering if using Neural network in which the last layer has Softmax activation with negative-log-likelihood as loss function and perform gradient descent also assumes IIA?

If so; is there anyway of make neural net violates IIA ?

  • $\begingroup$ What is "the IIA"? $\endgroup$
    – Sycorax
    Oct 14, 2019 at 22:46
  • $\begingroup$ @Sycorax I've updated the post to include IIA $\endgroup$
    – ElleryL
    Oct 14, 2019 at 22:47


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.