- [[L2 Regularization]]
- [[L1 Regularization]]
- [[Dropout Regularization]]
- reduces model [[Bias and Variance|Variance]] / overfitting
- $\lambda$ is the regularization parameter
- you set this using the dev set
- higher $\lambda$ incentivizes model to mellow down all parameters
- so only really important/impactful parameters stay
- while insignificant, potentially overfitting minor parameter nudges are wiped out
- results in an overall simpler model