Loss function for NN training which doesn't penalize inside given margin

Is there any loss function for neural network training in regression problems which does not penalize inside a given margin from given tensor (e.g. one unit far from given tensor) and out of this margin penalize based on distance from the same given tensor?Something like this