to apply a symmetric, continuously differentiable, piecewise analytic transform which is quadratic near zero, but linear past some threshold, to a quantity, especially an error. To do so maintains smooth derivatives for optimization purposes, but is more robust against outliers than purely quadratic error norms;
'dude, I totally huberized that residual.'
Prices shown in USD.
Type your email address below to get our free Urban Word of the Day every morning!
Emails are sent from firstname.lastname@example.org. We'll never spam you.