# Factorizing convolutional filters into smaller kernels

Can we find small kernels that convolve with each other to give a target kernel

In this notebook I look at factorizing large kernels into successive smaller ones. In particular, I look at a specific $7\times 7$ image filter that I want to implement as a convolution of two $3 \times 3$ filters. I talk about nonlinear optimization and show how to do this kind of problem in pytorch.

Check it out here.