[DNN] RMS Prop

This is another way to speed up gradient descent.

RMS Prop stands for Root Mean Squared propagation.

• Suppose in this example, w indicates the horizontal movement and b indicates the vertical movement.
• We compute S_dw and S_db.
• if S_dw is relatively small and S_db is relatively large, we can place these in the denominator and update w and b accordingly.
• Then, we can diminish the movement of b and speed up w (horizontal movement)
• With RMSprop, we can possibly use larger learning rate to speed up the process even more.
  ※ In practice, the horizontal might be w1, w2, w3 and the vertial might be w4, w5, w6 and b.