According to Mercer’s Theorem, what property must a kernel function's Gram matrix possess to be valid for use in an SVM?
Show answer
Positive semi-definite
A kernel represents an inner product in a high-dimensional feature space. For the optimization problem to be convex and solvable, the kernel must satisfy Mercer's condition, ensuring the resulting Gram matrix has non-negative eigenvalues.
Also asked as:
- To satisfy Mercer’s Theorem for use in an SVM, what property must the Gram matrix of a kernel function have?