IntermediateTopic 58 of 104Engineering Mathematics
Singular value decomposition
Factorises any matrix into rotations and scaling; used for compression, noise reduction and pseudo-inverses.
In depth
SVD factorises A=UΣVᵀ with orthogonal U, V and diagonal Σ of singular values. Large singular values carry most of the information, so truncating gives the best low-rank approximation. It also gives the pseudo-inverse and condition number.
Key points
- Works for any matrix
- Condition number = σmax/σmin
- Used for image compression and PCA
ExampleKeeping the top 20 singular values can compress an image by 10× with little visible loss.
SVD reveals the most important directions in any data or operator, so it underpins noise reduction, compression and robust inversion.
What is Singular value decomposition?
Factorises any matrix into rotations and scaling; used for compression, noise reduction and pseudo-inverses.
Why is it important to understand singular value decomposition?
SVD reveals the most important directions in any data or operator, so it underpins noise reduction, compression and robust inversion.
What do the singular values mean?
They are the stretching factors along orthogonal directions. Large values carry the main information, and tiny ones often mostly represent noise.
Why is SVD more robust than eigen-decomposition?
It works for any rectangular matrix and always exists, while eigen-decomposition needs square matrices and can fail for defective ones.
How is Singular value decomposition used in practice?
Typical applications include compressing a large sensor dataset while keeping its main features, computing the pseudo-inverse of an ill-conditioned matrix and estimating the rank of a noisy measurement matrix.
What should I learn before Singular value decomposition?
Study "Least squares" first (topic 57 of 104 in Engineering Mathematics). It covers the groundwork this topic builds on.
What are the key points of Singular value decomposition?
Works for any matrix. Condition number = σmax/σmin. Used for image compression and PCA.
Can you give an example of Singular value decomposition?
Keeping the top 20 singular values can compress an image by 10× with little visible loss.
What comes after Singular value decomposition?
The next topic is "First-order ODEs", which builds directly on this one.
How difficult is Singular value decomposition?
It is classed as Intermediate in Engineering Mathematics (topic 58 of 104). It is an intermediate topic. Work through the beginner topics in Engineering Mathematics first.