Is it possible to use modulo instead of squaring in the cost function?

Question

Neural networks

I study neural networks, I decided to write from scratch a solution to the problem of recognizing numbers in C ++ without using third-party libraries (for a better understanding of the principle of neural networks). Reached the cost function. Almost all manuals say that you need to calculate the sum of the squared deviations and find such weights and thresholds that the sum is minimal.

So, question. Squaring is a rather complicated operation, at least much more complicated than a module. Can modulo be used instead of squaring for optimization purposes?

Answer 1

it is possible, but it may work worse , it all depends on the data and the task, probably it will only be possible to understand experimentally