1
$\begingroup$

One part of my learning algorithm contains Huber loss. I would like to optimize it using gradient decent method. Unfortunately I don't know, how can I determine the value $h$ for that. Here is definition of Huber loss.

Second question is that, what is the gradient of Huber loss ?

enter image description here

  • 0
    Hint for problem 1: you'd like the loss function to be continuous at $h$. For problem 2: What happens if you look at each of the two cases separately? – 2017-02-14

0 Answers 0