$t$ is a binary target variable equal to $0$ or $1$, and below are the binary distribution and the log-likelihood error function.


$E(W)=-\displaystyle\sum_{n=1}^N (t_n \ln y_n + (1-t_n) \ln (1-y_n))$

But how should I write down the error function if there is a misclassification problem on a training data set? The exact wording is:

There is a probability $\epsilon$ that the class label on a training data point has been incorrectly set.

It's said that the error function above is obtained when $\varepsilon$ = 0.

Would appreciate any help or hint.


For each sample $i$, let $m_i=T$ if there is a misclassification and $m_i=F$ otherwise. Then:


Then the case $m_i=T$ flips $y_i$ to $1-y_i$. Can you finish it from here?


Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.