Two papers: Why do deep convolutional networks generalize so poorly to small image transformations? par Kelvin A Diversity-Promoting Objective Function for Neural Conversation Models par Thomas