Why do we use squared loss instead of absolute loss?
One reason is because by squaring the loss you can magnify it which can help train the model.
Another reason is because absolute loss is not differentiable when equals 0. This can complicate optimization process.
