---
id: wiki-2026-0508-l2-regularization
title: L2 Regularization
category: 10_Wiki/Topics
status: duplicate
canonical_id: l1-and-l2-regularization
duplicate_of: "[[L1-and-L2-Regularization]]"
aliases: []
source_trust_level: A
confidence_score: 0.9
verification_status: redirected
tags: [duplicate, regularization, ridge, weight-decay]
last_reinforced: 2026-05-10
github_commit: pending
---

# L2 Regularization

> **이 문서는 [[L1-and-L2-Regularization]] 의 중복본입니다.** Canonical 문서로 redirect.

## 핵심 요약 (specialized aspects)
- L2 = Ridge = weight decay. Penalty: $\lambda\|\beta\|_2^2$. 모든 계수를 0 근처로 수축하지만 정확히 0은 아님 — feature selection 효과 없음.
- Closed-form: $\hat\beta = (X^TX + \lambda I)^{-1}X^Ty$.
- Bayesian view: Gaussian prior on weights.
- Deep learning에서 SGD weight_decay 파라미터로 자주 사용. AdamW (decoupled weight decay)가 표준.
- L1과의 차이/조합 (Elastic Net)은 canonical [[L1-and-L2-Regularization]] 문서에 통합.

## 🔗 Graph
- 부모: [[L1-and-L2-Regularization]] (canonical)

## 🕓 변경 이력
| 날짜 | 변경 |
|---|---|
| 2026-05-08 | Phase 1 |
| 2026-05-10 | 중복 처리 — canonical 문서로 redirect |