Rumelhart, Hinton, and Williams published 'Learning representations by back-propagating errors' in Nature, popularizing the backpropagation algorithm for training multi-layer neural networks. Though the underlying idea had earlier roots, this paper made it central to the field and enabled networks to learn useful internal representations. A foundational moment for the deep-learning revolution decades later.