Neuron dropout is a deep learning regularisation technique that removes neurons by multiplying their activation output by 0, with some probability . This effectively trains sub-networks without certain neurons.