How is SVD related to PCA?
Updated May 16, 2026
Short answer
SVD decomposes data matrix and is numerically more stable than eigen decomposition in PCA.
Deep explanation
PCA can be computed using SVD on centered data matrix X = UΣVᵀ, where V contains principal components and Σ relates to variance.
Real-world example
Used in large-scale recommender systems for matrix factorization.
Common mistakes
- Confusing left and right singular vectors.
Follow-up questions
- Why is SVD preferred in practice?
- What does Σ represent?