Explaining the effects of non-convergent sampling in the training of energy-based models* | Publicación