• Gaussian prior on weights of a GLM lead to Ridge / $\ell_2$ squared regularization.
  • Laplace prior leads to $\ell_1$ regularization


What prior would lead to $\ell_\infty$ regularization ?


1 Answer 1


Thinking of similar examples, the reason behind the correspondence between these priors and norms is the exponent part. $$f(\beta)\propto \exp(-\lambda||\beta||_1)\rightarrow \ell_1$$ $$f(\beta)\propto \exp(-\lambda||\beta||^2_2)\rightarrow \ell_2 \text{ squared}$$ So, if we have a prior in the form below, it will correspond to $\ell_\infty$ because the negative log likelihood will directly include the term $\lambda||\beta||_{\infty}$ as a summand. $$f(\beta)\propto \exp(-\lambda||\beta||_\infty)\rightarrow \ell_\infty$$

I don't know if this PDF has a name, but it is a proper one ($n$ is the dimension): $$f(\beta)=\frac{\lambda^n}{2^nn!}\exp(-\lambda||\beta||_\infty)$$

  • 1
    $\begingroup$ Great answer. Thanks. $\endgroup$
    – dohmatob
    Jan 7, 2020 at 20:16

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.