Huber location estimator

The Huber location estimator solves ${\sum }_i{\psi }_k(x_i-\theta )=0$, where ${\psi }_k$ clips its argument to $[-k,k]$.

The Huber score clips a residual at a fixed threshold:

For standard normal distribution location estimation, an influence function $h$ obeys $\mathbb{E}[Zh(Z)]=1$. Under $|h|\le C$, minimize $\mathbb{E}{h}^2$ by projecting $ax$ onto $[-C,C]$, choosing $a$ to meet that identity. For $C>\sqrt{\pi /2}$ this gives the displayed Huber score normalized by $2\Phi (K)-1$, with $C=K/(2\Phi (K)-1)$. Pointwise minimality of $h^2-2axh$ proves global optimality after integration. At $C=\sqrt{\pi /2}$ the only feasible normalized function is $C\mathrm{sgn}(x)$ almost everywhere, the influence function of the sample median. The endpoint is a limit of rescaled clipped scores, not the identically zero score obtained by substituting $K=0$ literally.

The Huber loss is quadratic near zero and linear beyond a fixed cutoff; its derivative is the clipped Huber score.

A Huber loss applied to a discrete gradient combines quadratic penalization of small slopes with linear penalization of large slopes. This interpolates between squared-gradient smoothing and total variation denoising, allowing smooth regions while retaining an edge-preserving large-gradient regime. The threshold should be chosen consistently with the signal's units and discretization. It can reduce staircasing in total variation denoising without guaranteeing its elimination.