midPCA
What is the role of SVD in PCA?
Updated May 17, 2026
Short answer
PCA can be computed using Singular Value Decomposition (SVD) instead of covariance matrix.
Deep explanation
SVD decomposes data matrix into U, S, V matrices. PCA uses right singular vectors (V) as principal components. SVD is numerically more stable than covariance eigen decomposition.
Real-world example
Large-scale NLP embeddings reduction.
Common mistakes
- Assuming covariance method is always used.
Follow-up questions
- Why is SVD stable?
- Which is faster?