Rylan Schaeffer

Logo
Resume
Research
Learning
Blog
Teaching
Jokes
Kernel Papers


Linear Regression

Ordinary Linear Regression

In ordinary linear regression a.k.a. ordinary least-squares regression, we consider a linear model

\[\hat{y} = x^T \beta\]

and loss function

\[L_{OLS}(\beta) := \frac{1}{2N} \sum_n \lvert \lvert y_n - x_n^T \beta \lvert \lvert_2^2\]

The parameters that minimize the loss are

\[\beta^* = (X^T X)^{-1} X^T Y\]

Ridge Regression

In ridge regression, we consider a regularized loss function

\[L_{Ridge}(\beta) := \frac{1}{2N} \sum_n \lvert \lvert y_n - x_n^T \beta \lvert \lvert_2^2 + \lambda \lvert \lvert \beta \lvert \lvert_2^2\]

The parameters that minimize

Hat Matrix