Assume that we have a matrix $X$ and a weight matrix $W$. We multiply $W*X$ and get another matrix $Y$
$$Y = W*X$$
Now we are using support vector machine to see how separable the matrix $Y$ is. Assume that we get a result of $95$% separation. That menas $5%$ error.
Assuming that I have $W_{old}$ and its old accuracy number from support vector machine. The goal is to have $Y$ as much linear separable as possible and the sum of $W$ close to $0$ as much as possible.
Can I with that information know how much I should change the new weight matrix $W$?