CLOSURE: Assessing Systematic Generalization of CLEVR Models

Dzmitry Bahdanau 123 Harm de Vries 2 Timothy J. O'Donnell 14 Shikhar Murty 5 Philippe Beaudoin 2 Y oshua Bengio 136 Aaron Courville 136 1 Mila, Quebec Artificial Intelligence Institute 2 Element AI 3 Universit e de Montr eal 4 McGill University 5 Stanford University 6 CIFAR Fellow Abstract The CLEVR dataset of natural-looking questions about 3D-rendered scenes has recently received much attention from the research community. A number of models have been proposed for this task, many of which achieved very high accuracies of around 97-99%. In this work, we study how systematic the generalization of such models is, that is to which extent they are capable of handling novel combinations of known linguistic constructs. To this end, we test models' understanding of referring expressions based on matching object properties (such as e.g. "the object that is the same size as the red ball") in novel contexts. Our experiments on the thereby constructed CLOSURE benchmark show that state-of-the-art models often do not exhibit systematicity after being trained on CLEVR. Surprisingly, we find that an explicitly compositional Neural Module Network model also generalizes badly on CLOSURE, even when it has access to the ground-truth programs at test time. We improve the NMN's systematic generalization by developing a novel V ector-NMN module architecture with vector-valued inputs and outputs. Lastly, we investigate the extent to which few-shot transfer learning can help models that are pretrained on CLEVR to adapt to CLOSURE. Our few-shot learning experiments contrast the adaptation behavior of the models with intermediate discrete programs with that of the end-to-end continuous models. 1 Introduction The ability to communicate in natural language and ground it effectively into our rich unstructured 3D reality is a crucial skill that we expect from artificial agents of the future. A popular task to benchmark progress towards this goal is Visual Question Answering (VQA), in which one must give a (typically short) answer to a question about the content of an image.