What causes the increase in Gradient and Total Error under Stochastic Gradient Descet mode?

Question

I am trying to understand the intuition behind how neural networks learn. I understand the math behind it and have already tried to solve it analytically. While coding