1 result for “training stability”
Layer normalisation is a technique that normalises the inputs across the features of a single training example, stabilising and accelerating the training of deep neural networks, especially transformers.