For linear regression there's an assumption that error terms come from normal distribution. so that $Y = aX + b + \epsilon$, where $\epsilon$ has normal distribution with mean zero and certain variance.

Is there an analogous assumption for Huber loss (or MAE which should be easier)? So that for a variable described by relation $Y = aX + b + \epsilon$ with $\epsilon$ from this distribution huber regression (or theoretically minimization of MAE) gives a and b parameters?


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.