What is Back Propagation and Explain it’s Working.?

What is Back Propagation and Explain it’s Working.?


Backpropagation is a special case of gradient descent. I will focus on gradient descent in the following.

Imagine a surface with hills and valleys. You want to find the lowest point. You start at a random point. Then you look around you and choose the direction with steepest descent. You go a few steps (the learning rate) in that direction. Then you look again around you.

Yes, it is really that simple. But there are many variations. One common variation is called momentum. This is related to the physical value. You build up momentum by going in the same direction for some time.