Punishment is reducing a behaviour due to bad outcome e.g. A cow stops touching an electric fence because it gets a shock.
Negative reinforcement is increasing a behaviour that reduces a bad outcome. e.g. Put on a seat belt to stop the car bleating.
In this context, how are neural nets trained by Reinforcement Learning (RL) to approximate pi(s) and Q(s,a) ? Must negative neuron activations be allowed in hidden layers to enable negative reinforcement ? Are there practical issues with value ranges ?