Vanishing/Exploding Gradients - How can we prevent vanishing and exploding gradient?

When training a Neural Network, it is possible to prevent the disappearing and exploding gradients by properly initializing the weights. This approach limits the out-of-control outputs that occur during forward feedback as a result of layer activation. It may take more time for the network to converge if the loss function gradient becomes too tiny (vanishing gradients) or too big (exploding gradients), respectively.