Continual backpropagation is a technique that selectively reinitializes low-utility units in a neural network. It uses a utility measure called contribution utility, which is based on the value of connections and weights in the network. This measure is updated over ...