Trjtdtk

Can any one tell me what the effects of $L_2$ loss and smooth $L_1$ loss (i.e. Huber loss with $alpha = 1$) are, and when to use each of them ?

First, Huber loss only works in one-dimension as it requires $$left|boldsymbolaright|_2=left|boldsymbolaright|_1=delta$$at the intersection of two functions, which only holds in one-dimension. Norms $L_2$ and $L_1$ are defined for vectors. Therefore, in my opinion, Huber loss better be compared with squared loss rather than $L_2$ loss, since "$L_2$" presumes a multi-dimensional input compared to "squared".

Huber loss is the same as squared loss for differences less than $delta$, and the same as absolute loss for differences larger than $delta$, i.e.
$$beginalign*
L_delta(y_n, f_theta(boldsymbolx_n))
=left{
beginmatrix
frac12left(y_n - f_theta(boldsymbolx_n)right)^2 & left|y_n - f(boldsymbolx_n)right| leq delta,\
deltaleft|y_n - f_theta(boldsymbolx_n)right| - frac12delta^2, & textotherwise.
endmatrix
right.
endalign*$$

where $y_n$ is the target of data point $n$, and $f_theta(boldsymbolx_n)$ is model's prediction. Note that $L_delta$ has nothing to do with $L_p$ norm, despite the similar notations.

Because of this definition, for large differences due to outliers, gradient of loss function remains constant $pm delta$, the same as absolute loss, i.e.
$$fracy_n - f_theta(boldsymbolx_n)rightpartial theta_i = pm delta fracpartial f_theta(boldsymbolx_n)partial theta_i$$
compared to squared loss, where gradient increases with the difference, i.e.
$$fracpartial frac12left(y_n - f_theta(boldsymbolx_n)right)^2partial theta_i = -left(y_n - f_theta(boldsymbolx_n)right)fracpartial f_theta(boldsymbolx_n)partial theta_i$$

which leads to large contributions from outliers when we update a parameter solely based on squared loss as follows:
$$beginalign*
theta'_i &=theta_i + lambda sum_n fracpartial f_theta(boldsymbolx_n)partial theta_ileft(y_n - f_theta(boldsymbolx_n)right) \
&= theta_i + lambdasum_n notin textoutliers fracpartial f_theta(boldsymbolx_n)partial theta_i(textsmall) +lambdasum_n in textoutliers fracpartial f_theta(boldsymbolx_n)partial theta_i(textlarge)
endalign*$$

It is worth noting that, here, outliers are irregularities in the joint input-output space $(boldsymbolx_n, y_n)$, not necessarily just in the input space $boldsymbolx_n$ as we usually visualize in unsupervised tasks. For example, in a linear trend, none of $(x, y)=(1, 2)$, $(5, 10)$, $(10, 20)$ are outliers, but $(1, 10)$ is, which leads to large difference $(10 - 2)$ when model expects (predicts) $f_theta(1)=2$.

When to use each of them?

Reminding that we are only talking about one-dimensional targets, Huber loss is a complete replacement for squared loss to deal with outliers. However, the challenge is the choice of $delta$, which makes it a less favorable "first choice" when we are not yet familiar with the problem at hand. Therefore, we may start with squared loss (or other losses), and after a while try to experiment with Huber loss for different values of $delta$.