Theory

In traditional linear regression, we try to recover a hidden model parameter $\vec w*$ with samples $(\vec x, y)$ of the form $y = \vec {w}^{*T} \vec x + \epsilon$, where $\epsilon$ is sampled from some noise distribution.