Definition
Plain language
A classic statistics method for fitting a line through data while gently discouraging overfitting.
As stated in the literature
A linear regression variant with an L2 penalty on coefficients, admitting a closed-form solution and well-conditioned even in ill-posed settings.
Why it matters: It's a workhorse method whenever you have more features than data points, or strongly correlated inputs that would otherwise make plain regression blow up.
For example, when predicting house prices from a hundred correlated features, ridge regression pulls the noisy individual coefficients toward zero so the overall fit is more stable.
Heard on the show
“Specifically, ridge regression.”Episode 033 — Echo: The Paper Arguing You Never Needed a KV Cache for Retrieval