Let $A$ be a symmetric positive semi definite matrix, $A \in SPD^n$, and $K \in \mathbb{R}^{k\times k}$ a kernel.

What are the necessary and sufficient conditions over $K$ that will make the convolution of $A$ and $K$ $$ (A * K)_{ij} = \sum_u \sum_v A_{uv}K_{i-u,j-v} $$ Positive semi definite ?

  • From the numerical tests I ran, a PSD kernel seems to work. I can't figure out how to prove it though !
  • $\begingroup$ Do you convolve on the matrix? You can probably express it with Kronecker products in a higher dimension (vectorized) space. $\endgroup$ – mathreadler Aug 11 '17 at 20:28
  • $\begingroup$ @mathreadler I don't see how to express it as a Kronecker product. There is a way to express it as a Hadamard product which involves submatrices of $A$, but I don't find it useful. (Both products preserve PSD though). $\endgroup$ – Tool Aug 14 '17 at 17:49

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.