From NNDL chapter 3. explaining the equation will also help.

Problem 0 We've discussed at length the learning slowdown that can occur when output neurons saturate, in networks using the quadratic cost to train. Another factor that may inhibit learning is the presence of the :1: 3- term in Equation [61). Because of this term, when an input 1:3- is near to zero, the corresponding weight 1113- will learn slowly. Explain why it is not possible to eliminate the :L'j term through a clever choice of cost function